Syntactic Complexity: Diachrony, acquisition, neuro-cognition, evolution (Typological Studies in Language)

Syntactic Complexity Typological Studies in Language (TSL) A companion series to the journal Studies in Language. Vol...

Author: T. Givon | Masayoshi Shibatani

51 downloads 972 Views 5MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Syntactic Complexity

Typological Studies in Language (TSL) A companion series to the journal Studies in Language. Volumes in this series are functionally and typologically oriented, covering specific topics in language by collecting together data from a wide variety of languages and language typologies.

General Editor Michael Noonan

University of Wisconsin-Milwaukee

Assistant Editors Spike Gildea

University of Oregon

Suzanne Kemmer Rice University

Editorial Board Wallace Chafe

Matthew S. Dryer

Paul J. Hopper

Ronald W. Langacker

Doris L. Payne

Sandra A. Thompson

Bernard Comrie

John Haiman

Andrej A. Kibrik

Charles N. Li

Frans Plank

Dan I. Slobin

R.M.W. Dixon

Jerrold M. Sadock

Edith Moravcsik

Andrew Pawley

Bernd Heine

Santa Barbara San Diego

Leipzig / Santa Barbara Santa Barbara Melbourne Canberra

Buffalo

Oregon St Paul

Konstanz Chicago

Pittsburgh

Santa Barbara Moscow

Berkeley

Milwaukee

Köln

Volume 85 Syntactic Complexity. Diachrony, acquisition, neuro-cognition, evolution Edited by T. Givón and Masayoshi Shibatani

Syntactic Complexity Diachrony, acquisition, neuro-cognition, evolution

Edited by

T. Givón University of Oregon

Masayoshi Shibatani Rice University

John Benjamins Publishing Company Amsterdam / Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Syntactic complexity : diachrony, acquisition, neuro-cognition, evolution / edited by T. Givón and Masayoshi Shibatani. p. cm. (Typological Studies in Language, issn 0167-7373 ; v. 85) Includes bibliographical references and index. 1. Grammar, Comparative and general--Syntax. I. Givón, T. II. Shibatani, Masayoshi. P291.S952

2009

415--dc22 2008053096 isbn 978 90 272 2999 1 (hb; alk. paper) / 978 90 272 3000 3 (pb; alk. paper) isbn 978 90 272 9014 4 (eb)

© 2009 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents

Introduction T. Givón

1

part i. Diachrony From nominal to clausal morphosyntax: Complexity via expansion Bernd Heine

23

Re(e)volving complexity: Adding intonation Marianne Mithun

53

Multiple routes to clause union: The diachrony of complex verb phrases T. Givón

81

On the origins of serial verb constructions in Kalam Andrew Pawley A quantitative approach to the development of complex predicates: The case of Swedish Pseudo-Coordination with sitta “sit” Martin Hilpert & Christian Koops

119

145

Elements of complex structures, where recursion isn’t: The case of relativization 163 Masayoshi Shibatani Nominalization and the origin of subordination Guy Deutscher The co-evolution of syntactic and pragmatic complexity: Diachronic and cross-linguistic aspects of pseudoclefts Christian Koops & Martin Hilpert Two pathways of grammatical evolution Östen Dahl

199

215 239

 Table of contents

part ii. Child language On the role of frequency and similarity in the acquisition of subject and non-subject relative clauses Holger Diessel ‘Starting small’ effects in the acquisition of early relative constructions in Spanish Cecilia Rojas-Nieto The ontogeny of complex verb phrases: How children learn to negotiate fact and desire T. Givón

251 277

311

part iii. Cognition and neurology Syntactic complexity versus concatenation in a verbal production task Marjorie Barker & Eric Pederson

391

The emergence of linguistic complexity Brian MacWhinney

405

Cognitive and neural underpinnings of syntactic complexity Diego Fernandez-Duque

433

Neural mechanisms of recursive processing in cognitive and linguistic complexity Don M. Tucker, Phan Luu & Catherine Poulsen

461

Syntactic complexity in the brain Angela D. Friederici & Jens Brauer

491

part iv. Biology and evolution Neural plasticity: The driving force underlying the complexity of the brain Nathan Tublitz

509

Recursion: Core of complexity or artifact of analysis? Derek Bickerton

531

Index

545

Introduction T. Givón

University of Oregon

1. Complexity and syntax Complexity is a property of organized entities, of organisms, or of systems. Individual entities, as long as they have no internal organization, are by definition maximally simple. They may become parts of an organized whole when they have some relations — or connectivity — within an organized system. Within an organized system, the simple entities may bear relations to the system as a whole, or to its subparts, or to each other. At the most abstract level, a system may be described as a network of nodes and connections, where the nodes stand for either the simplest entities or to more abstract, higher level sub-parts of the system, and the connections stand for the nodes’ relations within the system. As an illustration, consider (1a) below, which depicts ten simple entities with no connection to each other, thus not part of an organized system. Contrast (1a) now with (1b), where four of the simple entities partake in an organized system of the most rudimentary sort. In such a system, the individual nodes bear relations only to the system as a whole, which is then the sole abstract node.

(1) a. b.

The 4-node system in (1b) may be also depicted, using a tree diagram, as a 1-level hierarchy, with only one abstract node — or meta-level — representing the whole. That is:

(2) 1-level hierarchic system:



T. Givón

System (2) has the lowest complexity level, a 1-level hierarchy. But the same four entities may be re-organized in a more complex, 2-level hierarchy, as in:

(3) 2-level hierarchic system:

The gist of Herbert Simon’s (1962) seminal work on the architecture of complexity is that increased complexity is, at the most general level, increased hierarchic organization; that is, an increase in the number of hierarchic levels within a system. And as Chomsky (1957) has made explicit, the syntax of a simple transitive clause can already be a 3-level hierarchy:1

(4) Simple transitive clause: 3-level hierarchy S NP ADJ

VP NP

V

N

ADJ

smart

women

N

like smart

men

Two or more simple clauses may be concatenated into a clause chain, as in:

(5) Concatenated (conjoined, chained) clauses: C S NP N

CONJ VP

BE

S NP

ADJ

PRO

VP NP

V ADJ

women

are

smart and

they

like smart

N men

. If the lexical words themselves are considered the terminal nodes, then (4) is, trivially, a 4-level hierarchy. But since there is no hierarchic branching below the lexical category node (N, V, ADJ), a more realistic view is that lexical category nodes are the terminal node in a 3-level hierarchic structure.

Introduction

While chained clauses are definitely an iterative structure, whether they are embedded (‘recursive’) may be a matter of definition. Thus Chomsky (1957, 1965) considered them embedded under a higher node of the same type (S), thus presumably ‘recursive’. That is:

(6) Clause chaining (conjunction): S

S

S

S

S

While one may have good reasons for not considering chained clauses embedded, as in (6), they still figure prominently in the genesis of syntactic complexity. The type of syntactic complexity that concerns us most in this volume involves the embedding of clauses inside other clauses. The two most common types of such embedding are relative clauses in the noun phrase and verbal complements in the verb phrase, as in (7a) and (7b) below. Respectively:2

(7) a. Embedding in the NP (REL-clause) S NP

VP S/REL

N NP

ADJ

are

smart

VP V

PRO

women

BE

NP ADJ

[who like smart

N men]

. While REL-clauses and V-complements are the most common types of recursive embedding in syntax, they are by no means the only ones. And recursive syntactic structure may also involve constituents other than the clause, e.g., noun-phrase conjunction, as in: S SUBJ [NP] NP conj NP

VP V

OBJ [NP] NP conj NP

Jill and Jane love Joe and Jim





T. Givón

b. Embedding in the VP (V-complement) S NP ADJ

VP N

V

S/COMP NP

VP NP

V

smart

women think

PRO

ADJ

[they

like smart

N men]

Embedding a clause inside another clause is what Chomsky (1957, 1965) called recursivity; that is, a node at a lower hierarchic level being dominated by a node of the same type at a higher level. In a recent joint paper (Hauser et al. 2002), Chomsky and his co-authors assert that recursivity is the key feature that separates human language from pre-human communication. It is doubtful that such a statement could have made the splash it has made but for Chomsky’s name being attached to it. For some rather remarkable claims were advanced in that joint paper, ostensibly about the evolution of grammar. To wit: i. The evolution of recursivity was not subject to adaptive selection. ii. The evolution of recursivity proceeded non-gradually and without preadaptations. iii. Recursivity, a single abstract feature, is the only significant universal of human grammar, as compared to pre-human communication. iv. By inference, what is universal in language and biology, are the extant features of a system, rather than the general principles that govern the development of the system, be it evolutionary, ontogenetic or diachronic. v. Pre-human organisms (and human pre-language) do not display recursivity. Each one of these claims is open to obvious challenges. Thus, claims (i), (ii) and (iv) are sharp departures from the standard Darwinian view of biological evolution. Claim (iii) runs in the face of the extreme richness of human-specific grammar(s) as known to descriptive linguists. And claim (v) is a denial of wellknown recursive organization in visual, motor, memory, neural and communicative

Introduction

systems — and corresponding behaviors — of pre-human organisms and pre-linguistic humans.3 Finally, claim (iv) is also a sharp departure from Chomsky’s own view of universals of grammar.4 At this juncture there is no need to review in full detail Chomsky’s approach to the evolution of recursivity, or of language. A detailed critique may be found in Bickerton (this volume; forthcoming). Chomsky’s construction-bound notion of recursivity was tacitly granted by Everett (2005), even as he purported to attack it. Everett suggested that at least one language (Piraha, an Amazonian isolate) had no embedded REL-clauses or V-complements, but performs their communicative function using paratactic functional equivalents. This claim hinges on the diachronic rise of embedded clauses out of paratactic precursors, as well as on intonational packaging of clauses. We will deal with this issue in more detail elsewhere (Givón 2008, ch. 4,5).

2. Developmental domains The genesis of syntactic complexity, and indeed of human language, may be probed in three developmental domains: • Diachrony (historical change) •• Ontogeny (language acquisition) •• Phylogeny (evolution) Of these three, extensive data are available on the first two. The third is obviously more problematic, given the near-absence of data on the communication of

. Bird songs; Chimpanzee drumming-and-shrieking signals; lions, wolves and chimpanzee group hunting routines; chimpanzee tool-making routines — are all striking examples of complex, hierarchic and often recursive organization of behavior, communicative or other. Though of course, in non-human communication there is no isomorphic mapping between the complex syntactic signal and corresponding message units, a mapping so characteristic of human communication. That is, complex animal signals carry, mostly, simple holistic messages (‘come hither’, ‘back off ’, ‘food’, ‘predator!’, etc.). . In a discussion of his principles and parameters format, Chomsky (1992) says: “… [early generative grammar assumed that] each language is a rich and intricate system of rules that are, typically, construction particular and language particular … The principles-and-parameters approach that has developed in recent years, and that I assume here, breaks radically with that tradition … The notion of grammatical construction is eliminated, and with it, constructionparticular rules. Constructions such as verb phrase, relative clause, passive etc. are taken to be taxonomic artifacts, collections of phenomena explained through the interaction of principles of UG, with the values of parameter fixed …” (1992: p. 3; bracketed material and emphases added). Recursivity is, of course, a construction-based notion.





T. Givón

hominids during the protracted period — 6–7 million year — of the separation from our nearest great-apes relatives. To some people, this paucity of data on the intermediate stages of language evolution suggests a gapped evolution (Chomsky 1968; Hauser et al. 2002). To others it suggests that any discussion of the gradual evolution of human language is speculative and pointless (Slobin 2002). The approach pursued here harkens back to Givón (1979), Bickerton (1981) and Heine & Kuteva (2007), in assuming that one can gain valid insights about language evolution by inspecting the data from the two other developmental trends, diachrony and ontogeny, as well as from comparative pre-human communication and the study of pidginization, creolization and neurology. Citing the data of language diachrony and language ontogeny in order to argue for parallel trends in language evolution is of course not free of problems. One has to argue for purported common trends, and then decide whether they are only analogical and suggestive, or perhaps homologs with shared mechanisms. One must also note that similarities, however striking, are never absolute. Still, such similarities do exist, and are indispensable for our understanding of language evolution; especially that the alternative is not all that attractive — giving up on a theoretical account of language evolution. Put another way, in science any hypothesis with empirical consequences is preferable to no hypothesis. For the former may be empirically falsified, while the latter is impervious to empirical testing.

3. Developmental trends in the genesis of syntactic complexity 3.1 From words to simple clauses As noted earlier above, the syntax of simple clauses is already hierarchic, thus complex. The genesis of simple clauses in both ontogeny and evolution is of considerable interest on its own merit. At least theoretically, one could go in two directions: •• Expansion (analysis): Start from an undifferentiated holistic signal and then re-analyze it as a composite of morphemes (a word), or of words (a clause). •• Combination (synthesis): Start from the small units (morphemes, words) and then build up the word or clause combinatorially. In a recent paper, Tallerman (2007) made a convincing argument that the holistic-to-analytic expansion model is implausible for language evolution. Given the data of both diachronic syntax (Givón 2008, ch. 4,5) and child language acquisition (Givón 2008, ch. 7,8), Tallerman’s argument holds equally well for the two other developmental domains — at least in the domain of syntax. Thus regarding

Introduction

diachronic syntax, Heine & Kuteva (2007) have argued that the holistic-to-analytic trend is unattested: “… For example, there is little evidence in attested morpho-syntactic changes for hypotheses that have been proposed on the basis of what may be called the holistic (or analytic) hypothesis: On this hypothesis, early language was characterized by holistic, monomorphemic linguistic signals conveying propositional contents, and these signals are believed to have undergone a segmentation process whereby complex but unanalyzable signals were broken down into words and syntactic structures (Wray 1998, 2000; Kirby 2002; Arbib 2005: 119). While we concur with the proponents of this hypothesis (e.g., Callahan 2006) that pragmatics is a crucial factor in language change, we are not aware of any diachronic evidence to the effect that such a segmentation process can commonly be found in language change. New grammatical categories do not normally arise via the reinterpretation of complex, unanalyzed propositions; accordingly, we consider this hypothesis to be less convincing for reconstructing language evolution …” (2007: pp. 25–26)

In first-language acquisition, the child’s early holistic expressions are not clause-size utterances that code a holistic message, but single-word utterances that stand for a clausal message in a highly specific context, while having a clear lexical meaning; that is, the so-called one-word stage (Bloom 1973; Scollon 1976; see Givón, 2008, ch. 10). Children build up the components of one-word clauses later on, not by re-analzing the single word into its sub-parts, but by adding the missing words through the two-word stage and onward (Bowerman 1973). In this process, the lexical meaning of the word, already established during the one-word stage, does not change. Rather, during the one-word stage the rest of the propositional message is inferred from the context, in close collaboration with the adult interlocutor and in supplementation by the surrounding adult turns (Ervin-Tripp 1970; Scollon 1976; Ochs et al. 1979; see Givón 2008, ch. 7,8,10). There is one area in child language acquisition where the holistic-to-expansion (analysis) model has been well documented — the acquisition of morphologicallycomplex words and, likewise, of multi-word collocation on their way to becoming complex words; that is, what MacWhinney (1982) calls ‘rote learning’. But this is not the template for the rise of syntactic complexity by expansion and (re)analysis, only for the rise of some cases of morphological complexity (Tomasello 2000, 2003). In diachrony, the situation is altogether unambiguous. Grammatical morphology always arises from erstwhile lexical words; so that morphological complexity arises from syntactic complexity via combination — and later condensation — of syntactic constructions (Givón 1971, 1979; Heine et al. 1991; Heine & Kuteva 2007; Dahl in this volume; inter alia; see further discussion in Givón 2008, ch. 4,5).





T. Givón

3.2 From clauses to chains to embedded clauses There is strong cumulative evidence that the developmental trend in the genesis of syntactic complexity, in diachrony, ontogeny, and no doubt in evolution, is primarily compositional (synthesis), following the general trend: (8) General trend in the genesis of syntactic complexity: a. single words > simple clause b. simple clause > clause chains (parataxis) c. clause chains > complex/embedded clauses (syntaxis) In child language development, all four stages are well documented (Bloom 1973; Bowerman 1973; Scollon 1976; Ochs et al. 1979; see Givón 2008, ch. 7,8). In language diachrony, trends (7a) and (7b) are of course absent, since adults already use multipropositional discourse, (clause chains). But the last stage (7c) is well documented in diachrony, as is its direction — from parataxis to syntaxis (Heine & Kuteva 2007; Hilpert & Koops in this volume; Koops & Hilpert in this volume; Pawley in this volume). That is, composition (synthesis) rather than expansion (analysis) is the prevalent trend. And spontaneous animal communication seldom progresses beyond the single-word stage. (Tomasello & Call 1997; Savage-Rumbaugh et al. 1993; Pepperberg 1999). No direct evidence exists for intermediate stages of language evolution. However, cogent arguments can be raised for the proposition that child language and language diachrony, as well as pidginization and non-human communication, could furnish relevant evidence for general phylogenetic trends. Such general trends can be extracted from the works of Bloom (1973); Bowerman (1973); Bates (1976); Scollon (1976); Bickerton (1981, 1990); Heine & Kuteva (2007), or Givón (1979, 1989, 2005, 2008) inter alia. They include: (9) General developmental trends: a. Words before clauses b. one-word clauses before multi-word clauses c. Single-clause discourse before multi-clause discourse d. Chained clauses before subordinate/embedded clauses e. Nominal objects before clausal complements f. Single-word restrictive modifiers before clause-size modifiers g. Pre-grammatical (pidgin) communication before grammar h. Manipulative speech-acts before declaratives and interrogatives i. Deontic modality before epistemic modality j. Non-displace spatio-temporal reference before displaced reference These trends are discussed in considerable detail in several chapters below. The comparative study of non-human communication (Tomasello & Call 1997; Savage-Rumbaugh et al. 1993; Pepperberg 1999; Rumbaugh & Washburn 2003; Cheyney & Seyfarth 1990, 2007; Zuberbühler 2000, 2001, 2002; Crockford & Boesch

Introduction

2003, 2005; inter alia) is consonant with the general trends in (9), furnishing evidence for the early trends (9a,b,c). All this is not to say that the similarities between — and mutual relevance of — language diachrony and ontogeny and language evolution are a settled issue. Still, the similarities, and the shared trends (9a–j), furnish a broad theoretical and empirical context within which relevant questions may be raised. Some of those are: •• ontogeny: What is the interaction between genetic endowment and external input? •• diachrony: Is adult-initiated change, which is not genetically coded, still genetically constrained? And if so, how? •• diversity: Is cross-language diversity of human languages, seemingly mediated by diachrony, genetically constrained, and if so how? •• evolution: Can the cumulative goal-directed adaptive behaviors by individual organisms contribute to genetically-coded endowment? And if so, how? •• parallels: Are the parallels between the three developmental trends parallels in product, or in the mechanisms that yield the product? •• analogy: Are parallels between the three developmental trends merely analogical, or do they also involve homology? •• biology: What is the bio-genetic basis of developmental parallels?

4. The neuro-cognitive basis of syntactic complexity 4.1 Cognition and syntactic complexity Two related possible isomorphisms concern us here, one between language and mind, the other between mind and brain. As suggested elsewhere (Givón 2008, ch. 2), some mappings between language and mind are relatively transparent. That is, if one takes Atkinson and Shiffrin’s (1968) three-way division of the cognitive representation systems in (10) below, the translation into linguistic systems is relatively straightforward: (10) Cognitive representation systems:

System

Cognition

Language/grammar Units

a. semantic memory concepts b. episodic memory-I events/states c. episodic memory-II event chains

System

Units

lexical semantics propositional semantics discourse pragmatics

words clauses clause chains



 T. Givón

But where do complex-embedded clauses fit into this scheme? Grammar is used to code both levels (10b) and (10c) of episodic representation, and it does so by separating the task, roughly, into two parts: •• simple-clause grammar (event structure) •• complex-clause grammar (communicative intent) The latter part, the grammar of complex clauses, comprises the bulk of the morpho-syntactic machinery of human language — the grammar of subordinate, non-declarative, negative, de-transitive clauses (Givón 2001). This is the part of grammar Chomsky’s older formats (1957, 1965 ch. 3) treated as transformations. Such clause-types are much less frequent in natural discourse and more difficult to process (Givón 1995, ch. 2). With respect to such marked clause-types, our task is to look for neuro-cognitive differences between them and the unmarked simple/chained clause. Correlations between syntactic complexity as described by the linguists and cognitive complexity as described by the psychologist is not easy to articulate. Intuitively, one would like to assume at least two syntactic-cognitive correlations, which taken together yield a third. Respectively: (11) Possible correlations between linguistic and cognitive complexity: a. C oding: More complex mentally-represented events are coded by more complex linguistic/syntactic structures. b. Processing-I: More complex mentally-represented events require more complex mental processing operations. Therefore, c. P rocessing-II: More complex syntactic structures require more complex mental processing operations. To what degree such correlations are upheld by the experimental neuro-cognitive data remains to be assessed.

4.2 The neuro-cognition of syntactic complexity A number of neuro-cognitive experimental studies, all within the sentence processing tradition, suggest that relative clauses are harder to process than conjoined clauses (Just et al. 1996; Booth et al. 1999, 2000; Caplan et al. 2006a, 2006b, inter alia). But the very same studies also suggest that object REL-clauses are harder to process than subject REL-clauses. What is more, the increased brain activation, thus processing difficulty, between subject REL-clauses and object REL-clauses occurs in the same brain region (IFG, Broca’s area) as the increased activation between conjoined clauses and subject-REL-clauses. Put another way, the brain locus and increased activation are not specific to embedding, but rather to a more general factor of cognitive complexity. And further, several recent reviews of the experimental literature on syntactic complexity (Fernández-Duque this volume; Hagoort 2008; Kaan 2008) conclude that increased

Introduction

activation in the IFG may not be specific to syntactic complexity, perhaps not even to grammar or language. A second experimental tradition suggests an altogether different take on complexity. The processing of sequential information in perception, memory, and motor behavior involves chunking, whereby sequentially presented information that is longer than 3–4 items is re-coded into chunks of hierarchically organized units. Such chunking is strongly dependent on repeated exposure (frequency of access), so that habituated expert performers — typists, musicians, dancers, readers, chess players — organize and chunk their knowledge more hierarchically than novices (Chase & Simon 1973; Chase & Ericsson 1982; Schneider 1985; Gobet 2005; inter alia). What is more, chunked hierarchic structures are implicated in increased automaticity of processing, thus decreased mental effort and attentional demands (Posner & Snyder 1974; Schneider & Shiffrin 1977; Schneider 1985; Schneider & Chein 2003). The two traditions appear to make the opposite claim about mental processing difficulty, thus presumably cognitive complexity. The first purports to show that complex clauses are harder to process; the second that hierarchic structures are processed more efficiently. The resolution of this apparent contradiction is discussed elsewhere (Givón 2008, ch. 11).

4.3 Brain localization The old Geschwind (1970) model divided the load for language processing in the cortical left hemisphere between the pre-frontal Broca’s area, responsible for grammar, and the temporal Wernicke’s area, responsible for meaning. The two areas were said to interact through a dense neural connector, the arcuate fasciculus. An immense amount of work done since then has gradually elaborated on the pioneering lesion-based studies of Broca and Wernicke. This still-cumulating work has taken advantage of an array of sophisticated brain-imaging techniques that allowed much finer spatial resolution (PET, fMRI) and much finer temporal resolution (ERP). Both the pre-frontal and temporal sites, as well as the connecting channels, have been shown to have anatomically- and functionally-distinct sub-modules. Among those, the most relevant to our discussion of complexity are: a. A lexical-semantics circuit: Most likely a multi-modular circuit, with a core part perhaps located in the ventri-lateral pre-frontal cortex (pars orbitalis; BA 47/12); as well as one or more temporal sites in the posterior Superior Temporat Gyrus (STG) and Medial Temporal Gyrus (MTG), with some R-parietal and other components (Petersen et al. 1988; Raichle et al. 1993; Snyder et al. 1995; Posner & Pavese 1997; Abdulaev & Posner 1997; Bookheimer 2002; Badre & Wagner 2007;



 T. Givón

Martin & Chao 2001; inter alia); with the frontal-temporal connection via the extreme capsule (Schmahmann et al. 2007). b. A combinatorial propositional-semantics circuit: Most likely a multi-modular circuit of at least two sites, one prefrontal (pars opercularis; fOP; BA 44)and one temporal (anterior-STG), connected via the fasciculus uncinatus (Posner & Pavese 1997; Abdulaev & Posner 1997; Friederici & Frisch 2000; Friederici et al. 2006a, 2006b; Grodzinsky & Friederici 2006; Bahlmann et al. 2008; Friederici 2008; Pulvermüller 2002, 2003; Pulvermüller & Assadollahi 2007). c. A hierarchic/complex-clause circuit: Most likely a multi-modular circuit of two or more sites, one prefrontal (Broca’s area, BA 45a/45b), the other temporal (posterior STG), connected through the fasciculus longitudinalis superior (Friederici & Frisch 2000; Friederici et al. 2006a, 2006b; Grodzinsky & Friederici 2006; Bahlmann et al. 2008; Friederici 2008, Friederici and Brauer, in this volume). In addition, several sub-modules of the executive attention network (Schneider & Chein 2003; Posner & Fan 2008) must also be involved in grammatical processing, as well as working memory and episodic memory. And the cerebellum may also be involved (Argyropoulos 2007). For further discussion see Givón (2008, ch. 11).

5. Bio-evolutionary context While the brain is the proper biological site of the central processors of language, the evolution of general biological design is of some interest because it sets up general patterns and principles that may govern the rise of the brain mechanisms that underlie syntactic complexity. Some of these general developmental principles are: •• Graduality: By and large, the evolution of complex biological design is gradual, going from the more simple to the more complex. And there are both low-level biochemical and higher-level adaptive reasons for this graduality. At the biochemical level, major structural jumps are chemically unstable or enzymatically incompatible. At the adaptive level, major structural jumps cannot be accommodated by the current behavioral capacities of the organism. •• terminal modification: In conjunction with graduality, terminal modification reveals bio-evolution to be a tinkerer rather than an engineer. Existing structures are recruited to perform new but similar functions, simultaneously with the old functions. Eventually structural differentiation takes place, so that a new structure emerges adjacently to the old one, and now performs the new function. Eventual re-structuring and simplification may also take place, shrinking superfluous pieces of excess structure. The end result of this complex, protracted process

Introduction 

is, most commonly, that organs performing related functions are added at the end of the chain of a developmental sequence, near organs that perform older precursor functions. •• Adaptive motivation: By and large, the rise of new structures is guided by adaptive selection. What is more, the first stage involves the functional extension of an old structure, i.e., its recruitment to perform a related-but-not-identical function. The early stage of a biological innovation thus tends to involve functional ambiguity of old structures. Thus Mayr (1976) observes: “… the evolutionary changes that result from adaptive shifts are often initiated by a change in behavior, to be followed secondarily by a change in structure … The new habit often serves as the pacemaker that sets up the selection pressure that shifts the mean of the curve of structural variation …” (Mayr 1976: p. 106)

And further: “… Another entirely different and much more dramatic way by which evolutionary novelties can be acquired is a change in the function of a structure. Here an existing structure, let us say the [sensory] antenae of Daphnia, acquires the additional function of a swimming paddle and, under new selection pressures, becomes enlarged and modified … During a succession of functions, a structure always passes through a stage when it can simultaneously perform both jobs …” (Mayr 1997: p. 194–195)

•• Combination vs. expansion: As noted earlier above, Tallerman (2007) argues that combination (synthesis) — the joining together of smaller simple units into a larger complex one — rather than expansion (analysis) is the only plausible model for the evolution of syntactic complexity. In language diachrony, combination seems to be the general trend, especially in the genesis of morphological and syntactic construction (see further discussion in Givón 2008). In language ontogeny, various claims have been made about expansionfrom-holistic being the proper developmental model (Tomasello 2000, 2003; Diessel 2005). In his extensive survey, MacWhinney (1982) identified a cluster of instances of expansion-from-holistic (‘rote learning’) in child language acquisition. Without exception, they involve either bound morphology or complex words, a generalization that had often been obscured by conservative writing systems. In the two studies below on the ontogeny of complex constructions (Givón 2008, ch. 7,8), the developmental trend is, invariably, combination (synthesis). In bio-evolution, numerous examples of both combination (synthesis) and expansion (analysis) are attested. What is more, the two are sometime coupled within the same complex development. Examples of combination abound. Thus, the eukaryote

 T. Givón

protozoan cell is apparently an evolutionary combination of one primitive prokaryote cell that supplied the protoplasm and nucleic acids, with three smaller prokaryote cells, in a symbiotic arrangement, with the latter three contributing the mitochondria, plastids and flagella of the resulting eukaryote cell (Margulis 1981; Sleigh 1989; Yang et al. 1985). The incorporation of viruses into the DNA/RNA of many, perhaps most, organisms is another such example. Diploid cells with two sets of chromosomes, the current norm in higher organisms, are the combinatorial product — in both evolution and ontogeny (meiosis, sexual reproduction) — of two haploid precursor cells, each with only one set of chromosomes. And multi-cellular organisms in general began as temporarily-combined colonies of single-cell organisms. Mitosis (normal cell division) involves both expansion and combination. The first phase, the splitting of the diploid nuclear chromosomes into two haploid clusters, and then the copying of each haploid strand, is an example of expansion. But the subsequent re-combination of the two haploid nuclei is an example of combination. Likewise, the life-cycle of many organisms, most conspicuously sexual reproduction, involves a multi-cellular stage (combination) and a mono-cellular stage whose individual cells split from the larger organism (expansion). Likewise, in the growth of tissues and organs, cell division (expansion) is coupled with cell addition (combination). Higher up on the complexity scale, in the evolution and ontogeny of the nervous system, one could cite cases of expansion — the segmentation of the simple spinal chord; the projection of the cortex from the limbic-thalamic mid-brain. But one could also cite cases of combination, such as the joining together of erstwhile unrelated modules into a distributive network that now performs a complex higher function (executive attention, grammar, phonology). Why both the diachrony and ontogeny of syntactic complexity tilt heavily toward combination, and seem to shun expansion, is probably not due to general biological principles, but rather to more specific adaptively constraints.

•• Similar constraints at multiple levels of development: The evolution of the eye reveals the existence of similar developmental processes, supported by similar genetic structures, but applying at widely separated evolutionary stages. Thus, the eye of the squid and octopus is remarkably like that of vertebrates, but is nevertheless not its evolutionary homolog. In a paper discussing the evolution of social fish behavioral patterns, Fernald & White (2000) suggest that real-time adaptive behavior, ontogeny and phylogeny share a single mechanism, through which individual adaptive behavior influences genetic accretion: “… Behavior can and does influence specific aspects of brain structure and function over three different time frames. A causal link is easy to establish on an evolutionary time scale because selective forces of the ecological niche of the

Introduction

animal typically are reflected in the body shape, sensory and motor systems, and behavior. Similarly, on a developmental time scale, behavior acts in concert with the environment to establish structural changes in the brain that influence an organism throughout its lifetime. Surprisingly, there currently is evidence that in real time, social behavior also causes changes in the brain in adult animals …” (Fernald & White 2000: p. 1193; emphases added)

While the study in question concerns the adaptive behavior of social fish, its relevance to the three developmental trends discussed here is considerable. Real-time communicative behavior by individuals may be the common denominator of all three developmental trends in language — evolution, ontogeny and diachrony. In the first, behavior interacts with genetic mutations. In the second, with the genetically pre-set (‘innate’) developmental sequence. In the third, behavior does not interact directly with genes, but rather with the automating, habituating , skillacquiring learning brain, that in turn interacts with genetic mutations during selection (Hinton & Nowlan 1987; Givón 1989). And it is the real-time adaptive behavior by individual speakers during communication that is the engine of diachronic change. This is in line with the so-called Baldwin effect (Baldwin 1896), as well as with niche construction and genetic assimilation (Waddington 1942, 1953; Deacon 2008). To cite Ernst Mayr again: “… Many if not most acquisitions of new structures in the course of evolution can be ascribed to selectional forces by newly-acquired behaviors (Mayr 1960). Behavior, thus, plays an important role as the pace-maker of evolutionary change ….” (Mayr 1982: p. 612)

One objection to using the data of child language acquisition and language diachrony to construct hypotheses about language evolution has been that they represent starkly different evolutionary stages of the human brain, that of, respectively, immature modern humans, adult modern humans, and various pre-modern hominids (Slobin 2002). Since cases like the squid/human eye are common in biology (Tublitz this volume), and since the principles articulated by Baldwin (1898); Waddington (1942, 1953); Mayr (1982) and Fernald & White (2000) are widely attested in biology, perhaps this argument can be resolved on more substantive empirical and theoretical grounds. And perhaps, as Karl Popper (1934) said about scientific hypotheses, by their fruit thou shall know them.

References Abdulaev, Y. & Posner, M. 1997. Time-course of activating brain areas in generating verbal associations. Psychological Science 8(1).



 T. Givón Argyropoulos, G. 2007. The subcortical foundations of grammaticalization. Ms, University of Edinburgh, Language Evolution and Computation Research Unit. Badre, D. & Wagner, A.D. 2007. The left ventrolateral prefrontal cortex and the cognitive control of memory. Neuropsychologia 45. Bahlmann, J., Schubotz, R.I & Friederici, A. 2008. Hierarchical sequencing engages Broca’s area. NeuroImage 42(2). Bates, E. 1976. Language and Context: The Acquisition of Pragmatics. New York NY: Academic Press Bickerton, D. 1981. Roots of Language. Ann Arbor MI: Karoma. Bickerton, D. 1990. Language and Species. Chicago IL: University of Chicago Press. Bickerton, D. (in this volume). Recursion: Core of complexity or artifact of analysis. Bickerton, D. Forthcoming. Science’s compass points due south-south-east. Ms, University of Hawaii at Manoa, Honolulu, Hawaii. Bloom, L. 1973. One Word at a Time. The Hague: Mouton. Bookheimer, S. 2002. Functional MRI of language: New approaches to understanding the cortical organization of semantic processing. Ann. Rev. of Neuroscience 25. Booth, J.R., MacWhinney, B., Thulborn, K.R., Sacco, K., Voyvodic, J. & Feldman, H.M. 1999. Functional organization of activation patterns in children: Whole brain fMRI imaging during three different cognitive tasks. Prog. in Neuropharmacology & Biol. Psychiatry 33. Booth, J.R., MacWhinney, B. & Harasaki, Y. 2000. Developmental differences in visual and auditory processing of complex sentences. Child Development 71(4). Bowerman, M. 1973. Early Syntactic Development. Cambridge: CUP. Caplan, D., DeDe, G. & Michaud, J. 2006a. Task-independent and task-specific deficits in aphasia comprehension. Aphasiology 9/10/11. Caplan, D., Chen, E. & Watters, G. 2006b. Syntactic and thematic effects on BOLD signal association with comprehension and determination of plausibility of sentences with relative clauses. Ms, Massachusetts Geneal Hospital, Boston MA. Chase, W. & Simon, H. 1973. Perception in chess. Cognitive Psychology 4. Chase, W. & Ericsson, A. 1982. Skill and working memory. In The Psychology of Learning and Motivation, Vol. 16, G. Bower (Ed.), New York NY: Academic Press. Cheyney, D. & Seyfarth, R. 1990. How Monkeys See the World. Chicago IL: University of Chicago Press. Cheney, D. & Seyfarth, R. 2007. Baboon Metaphysics. Chicago IL: University of Chicago Press. Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: The MIT Press. Chomsky, N. 1968. Language and Mind. New York NY: Harcourt, Brace and World. Chomsky, N. 1992. A minimalist program for linguistic theory. MIT Occasional Papers in Linguistics 1. Cambridge MA: The MIT Press. Crockford, C. & Boesch, C. 2003. Context specific calls in wild chimpanzees, pan troglodytes verus: Analysis of barks. Animal Behavior 66. Crockford, C. & Boesch, C. 2005. Call combinations in wild chimpanzees. Behaviour 142. Dahl, Ö. (in this volume). Two pathways of grammatical evolution. Diessel, H. 2005. The Acquisition of Complex Sentences, Cambridge: CUP. Ervin-Tripp, S. 1970. Discourse agreement: How children answer questions. In Cognition and the Development of Language, J.R. Hayes (Ed.), New York NY: Wiley. Everett, D. 2005. Cultural constraints on grammar and cognition in Piraha: Another look at the design features of human language. Current Anthropology 46(4). Fernald, R.D. & White, S.A. 2000. Social control of brains: From behavior to genes. In M. Gazzaniga (Ed.), 2000.

Introduction 

Friederici, A. 2008. Brain circuits of syntax: From neuro-theoretical considerations to empirical tests. Ms presented at the Ernst Struengmann Forum on Biological Foundations and Origin of Syntax, Frankfurt, July 2008. Friederici, A. & Frisch, S. 2000. Verb-argument structure processing: The role of verb-specific and argument-specific information. J. of Memory and Language 43. Friederici, A., Bahlmann, J., Heim, S., Scubotz, R.I. & Anwander, A. 2006a. The brain differentiates human and non-human grammar: Functiional localization and structural connectivity. Proc. Nat. Acad. of Sci. USA 103. Friederici, A. & Brauer, J. (in this volume). Syntax complexity in the brain. Friederici, A., Fiebach, C.J., Schlesewsky, M., Bornkessel, I. & von Cramon, D.Y. 2006b. Processing linguistic complexity and grammaticality in the kleft frontal cortex. Cereb. Cortex 16. Gazzaniga, M. (Ed.). 2000. The New Cognitive Neuroscience, 2nd Edn. Cambridge MA: The MIT Press. Gernsbacher, M.A. 1990. Language Comprehension as Structure Building. Hillsdale NJ: Lawrence Erlbaum. Givón, T. 1971. Historical syntax and synchronic morphology: An archaeologist’s field trip. CLS 7, University of Chicago, Chicago Linguistics Society. Givón, T, 1979. On Understanding Grammar. New York NY: Academic Press. Givón, T. 1981. Logic vs. pragmatics, with human language as a referee: Towards an empirically viable epistemology. J. of Pragmatics 6(2). Givón, T. 1989. Mind Code and Context. Hillsdale NJ: Lawrence Erlbaum. Givón, T. 2005. Context as Other Minds. Amsterdam: John Benjamins. Givón, T. 2008. The Genesis of Syntactic Complexity. Amsterdam: J. Benjamins. Givón, T. & Malle, B.F. (Eds). 2002. The Evolution of Language out of Pre-Language [Typological Studies in Language 53]. Amsterdam: John Benjamins. Givón, T., Posner, M., Kellog, W. & Yee, P. 1985. The tracking of referents in discourse: Automated vs. attended processes [TR 85–3]. Institute of Cognitive and Decision Sciences, University of Oregon Grodzinsky, Y. & Friederici, A. 2006. Neuroimaging of syntax and syntactic processing, Current Opinion in Neurobiology, 16 Hauser, M., Chomsky, N. & Fitch, W.T. 2002. The faculty of language: What it is, who has it, how did itevole? Science 298: 1569–1579. Heine, B., Claudi, U. & Hunnemeyer, F. 1991. Grammaticalization: A Conceptual Framework. Chicago IL: University of Chicago Press. Heine, B. & Kuteva, T. 2007. The Genesis of Grammar. Oxford: OUP. Hilpert, M. & Koops, C. (in this volume). A quantitative approach to the development of complex predicates: The Swedish serial verb sitta ‘sit’. Just, M.A., Carpenter, P.A., Keller, T.A., Eddy, W.F. & Thulborn, K.R. 1996. Brain activation modulated by sentence comprehension. Science 274. Koops, C. & Hilpert, M. (in this volume). The co-evolution of syntactic and pragmatic complexity: Diachronic and cross-lingustics aspects of psudoclefts. MacWhinney, B. 1982. Basic syntactic processes. In Development, Vol. 1: Syntax and Semantics, S. Kuczaj (Ed.), Hillsdale NJ: Lawrence Erlbaum. Margulis, L. 1981. Symbiosis in Cell Evolution. San Francisco CA: Freeman. Mayr, E. 1976. Evolution and the Diversity of Life. Cambridge MA: Harvard University Press. Mayr, E. 1982. The Growth of Biological Thought. Cambridge MA: Harvard University Press.

 T. Givón Mayr, E. 1997. This is Biology. Cambridge MA: Harvard University Press. Neville. H. 1995. Developmental specificity in Neurocognitive development in Humans. In The Cognitive Neuroscience, M. Gazzaniga (Ed.), Cambridge MA: The MIT Press. Neville, H., Mills, D.L. & Lawson, D.S. 1992. Fractionating language: Different neural systems with different sensitive periods. Cortex 2(3). Ochs, E. & B. Schieffelin. (Eds). 1979. Developmental Pragmatics. New York NY: Academic Press. Ochs, E., Schieffelin, B. & Platt, M. 1979. Propositions across utterances and speakers. In E. Ochs & B. Shieffelin (Eds), 1979. Pawley, A. (in this volume). On the origins of serial verb constructions in Kalam. Pepperberg, I.M. 1999. The Alex Studies: Cognitive and Communicative Abilities of Grey Parrots. Cambridge MA: Harvard University Press. Posner, M. & Snyder, C.R.R. 1974. Attention and cognitive control. In Information Processing and Cognition: The Loyola Symposium, R.L. Solso (Ed.), Hillsdale NJ: Lawrence Erlbaum. Posner, M. & Pavese, A. 1997. Anatomy of word and sentence meaning. In Colloquium on Neuroimaging of Human Brain Functions, UC Irvine, M. Posner & M. Raichle (orgs). Ms, National Acad. of Science. Posner, M.I. & Fan, J. 2004/2008. Attention as an organ system In Neurobiology of Perception and Communication: From Synapses to Society (The IVth De Lange Conference), J. Pomerantz (Ed.), Cambridge: CUP. Popper, K. 1934 [1959]. The Logic of Scientific Discovery. New York NY: Harper and Row. Pulvermuller, F. 2002. A brain perspective on brain mechanisms: From discrete neural assemblies to serial order. Progress in Neurobiology 67. Pulvermuller, F. 2003. The Neuroscience of Language. Cambridge: CUP. Pulvermuller, F. & Assadollahi, R. 2007. Grammar or serial order?: Discrete combinatorial brain mechanisms reflected by the syntactic Mismatch Negativity. J. of Neuroscience 19: 971–980. Pulvermuller, F. & Shtyrov, Y. 2003. Automatic processing of grammar in the human brain as revealed by the mismatch negativity. NeuroImage 20. Pulvermuller, F., Styrov, Y. & Carlyon, R.P. 2008. Syntax as reflex: Neurophysiological evidence for early automaticity of grammatical processing. Brain and Language 104(3): 244–253. Rumbaugh, D.M. & Washbutn, D.A. 2003. Intellingence of Apes and Other Ratuonal Beings. New Haven CT: Yale University Press. Savage-Rumbaugh, S., J. Murphy, R.A. Sevcik, K.E. Brakke, S.L. Wiliams & D.M. Rumbaugh. 1993. Language Comprehension in Ape and Child, Monographs of the Society for Research in Child Development, Serial 233, Vol. 58, nos 3–4. Schmahmann, J.D., Pandya, D.N., Wang, R., Dai, G., D’Arcueil, H.E., de Crespigny, A.J. & Wedeen, V.J. 2007. Associate fibre pathways of the brain: Parallel observations from diffusion spectrum imaging and autoradiography Brain, 130. Schneider, W. 1985. Toward a model of attention and the development of automatic processing. In Attention and Performance, XI, M. Posner & O. Marin (Eds), Hillsdale NJ: Lawrence Erlbaum. Schneider, W. & Chein, M. 2003. Controlled and automated processing: Behavior, theory and biological mechanisms. Cognitive Science 27. Schneider, W. & Shiffrin, R. 1977. Controlled and automated human information processing, I: Detection, search and attention. Psych. Review 84. Scollon, R. 1976. Conversations with a One-Year-Old Child. Honolulu HI: University of Hawaii Press.

Introduction 

Shapiro, D.C.A. 1978. The Learning of Generalized Motor Programs. Ph.D. dissertation, UCLA. Shapiro, D.C.A. & Schmidt, R.A. 1980. The schema theory: Recent evidence and developmental implications. In The Development of Movement Control and Coordination, J.A.S. Kelso & J.E. Clark (Eds), New York NY: Wiley. Shtyrov, Y., Pulvermuller, F., Naatanen, R. & Ilmoniemi, R.J. 2003. Grammar processing outside the focus of attention: An MEG study. J. of Cognitive Neuroscience 15(8). Simon, H. 1962. The architecture of complexity. Proc. Amer. Phil. Soc. 106(6). Sleigh, M. 1989. Protozoa and other Protists. Cambridge: CUP. Slobin, D. 2002. Language evolution, acquisition and diachrony: Probing the parallels. In T. Givón & B.F. Malle (Eds), 2002. Tallerman, M. 2007. Analysing the analytic: Problems with holistic theories of proto-language. Ms, University of Durham. Thelen, E. 1984. Learning to walk: Ecological demands and phylogenetic constraints. Advances in Infant Research, Vol. 3. Norwood NJ: Ablex. Tomasello, M. 2000. Do young children have adult syntactic competence? Cognition 74. Tomasello, M. 2003. Constructing a Language: A Usage-Based Approach to Language Acquisition. Cambridge MA: Harvar University Press. Tomasello, M. & Call, J. 1997. Primate Cognition. Oxford: OUP. Yang, D., Oyaizu, Y., Oyaizu, H., Olsen, G.J. & Woese, C.R. 1985. Mitochondrial origins. Proc. Nat. Acad. of Sci. USA 82. Zuberbühler, K. 2000. Referential labeling in Diana monkeys. Animal Behavior 59. Zuberbühler, K. 2001. Predator-specific alarm calls inCampbell's monkeys. Cercopithecus campbelli, Behavioral Ecology and Sociobiology 50. Zuberbühler, K. 2002. A syntactic rule in forest monkey communication. Animal Behavior 63.

part i

Diachrony

From nominal to clausal morphosyntax Complexity via expansion Bernd Heine

Universität zu Köln The study of the rise of syntactic complexity, in particular of clause subordination and recursive language structures has more recently become the topic of intense discussion. The present paper builds on the reconstruction of grammatical evolution as proposed in Heine and Kuteva (2007) to present a scenario of how new forms of clause subordination may arise. Taking examples from attested cases of grammatical development as well as using evidence that has become available on grammaticalization in African languages, it is argued that there are two major pathways leading to the emergence of clause subordination: either via the integration of coordinate clauses or via the expansion of existing clauses. The concern of this paper is exclusively with the latter pathway.

1. Introduction As shown in Heine and Kuteva (2007), grammaticalization theory offers a tool to push linguistic reconstruction back to earlier phases of linguistic evolution, that is, to phases where human language or languages can be assumed to have been different in structure from what we find today. Grammaticalization is a process which concerns the development from lexical to grammatical forms and from grammatical to even more grammatical forms and constructions. Underlying this process there is a cognitive mechanism whereby concrete and salient concepts serve as vehicles or structural templates to conceptualize less concrete and less readily accessible concepts — with the effect that linguistic expressions for concrete concepts, such as physical objects or actions, are recruited to express more abstract concepts. Applying grammaticalization theory to a sizable body of data from over 500 languages across the world (e.g., Lehmann 1982; Heine & Reh 1984; Heine, Claudi & Hünnemeyer 1991; Hopper & Traugott 2003; Bybee, Perkins & Pagliuca 1994; see also Dahl 2004), Heine and Kuteva (2007) hypothesize that it is possible to reconstruct some major lines in the development of functional categories, and they propose the layering model of grammaticalization depicted in figure 1 (concerning the evidence on which reconstruction rests, see Heine & Kuteva 2007; chapter 2).

 Bernd Heine

I

NOUN

VERB

II

ADJECTIVE

III

IV

V

VI

ADVERB

DEM

PRN

AGR

DEF

PAS

ADP

REL

CPL

ASP

CAS

NEG

TNS

SBR

Figure 1. Six hypothesized layers of grammatical development (Heine & Kuteva 2007: 111). Abbreviation: AGR = agreement marks; ADP = prepositions and other adposition ASP = (verbal) aspect; CAS = case marker; CPL = complementizer ; DEF = marker of definiteness (“definite article”); DEM = demonstrative; NEG = negation marker; PAS = passive marker; PRN = pronoun; REL = relative clause marker; SBR = subordinating marker of adverbial clauses; TNS = tense marker. The dotted line indicates that there is only indirect evidence for reconstructing this development.

These authors go on to argue (Heine & Kuteva 2007; chapter 5) that there are crosslinguistically two main ways in which clause subordination may arise: Either via the integration of two independent sentences within one sentence, or via expansion, that is, the reinterpretation of a thing-like (nominal) participant as a propositional (clausal) participant.1 This is a strong claim, namely that clause subordination is historically derived from non-subordinate sentences.2

1. This terminology is taking from Diessel (2004), who uses them for two distinct kinds of strategies used in first language acquisition to develop complex sentences. Observing that in first language acquisition complex sentences appear later than simple sentences, he proposes the following generalization: “Thus, while complement and relative clauses evolve via clause expansion, adverbial and co-ordinate clauses develop through a process of clause integration” (Diesel 2004: 4). 2. The reader is referred to Givón (2002; 2008; in this volume), which discusses a much wider range of processes than we are able to cover here and provides a coherent syntactic account of these processes.

From nominal to clausal morphosyntax 

The concern of this chapter is exclusively with the process of clause expansion, which so far has received little attention in morphosyntactic reconstruction. For example, Hopper and Traugott (2003: 176) propose a cline of clause combining leading from parataxis to subordination; but their concern is only with clause integration; as we will see below, this is not the only way in which clause subordination arises. Similarly, what Dahl (this volume), following Givón (1979; 1995), calls the paratactic > syntactic construction schema is not within the scope of the present paper. Concerning a critical view on the integration hypothesis, see Deutscher (this volume). The chapter, which is a follow-up of Heine and Kuteva (2007), is meant to show how devices that first served to structure noun phrases come to assume functions of subordination. This, however, does not necessarily mean that there was no previous form of subordination; as Harris and Campbell (1995: 282ff.) rightly emphasize, the rise of a new form of subordination may simply mean that an existing form was either modified or replaced. In some language families, no subordinate structures can be reconstructed though; for example, no specific relative clause marking can be reconstructed for the Germanic languages. But this does not mean that in the relevant families there previously were no corresponding subordination structures. The chapter is based on a small survey of “nominal” complement clause constructions in languages across the world. What I have to say about complement clauses presumably applies as well to some types of relative and adverbial clauses, but more research is required on this issue (see Givón 2008, ch. 5 for a detailed treatment of relative clauses; see also Deutscher 2008). I will not, however, be concerned with the rise of recursive structures3 (for which see Heine & Kuteva 2007; chapter 5). The term “construction” has received a wide range of applications in the recent literature. I will use the term for linguistic phenomena (a) that combine a specific form with a specific meaning, (b) that combine more than one linguistic unit with one or more other units, and (c) whose meaning is non-compositional (i.e., is not identical to the sum of its parts). There are three main methods for reconstructing earlier morphosyntactic situations, namely (a) studying historical records of contiguous developmental stages, (b) analyzing synchronic variation of co-existing related constructions, and (c) internal reconstruction (Givón 2008, ch. 5). Historical information on grammatical change in the languages of the world is unfortunately scanty, and many of the reconstructions proposed are based on (b) and (c) supported by crosslinguistic generalizations on grammatical change, that is, by applying the methodology of grammaticalization theory to synchronic linguistic data, even if a number of the reconstructions are also supported by (a), i.e., by attested historical data (see Heine & Kuteva 2007; chapter 5).

3. As rightly pointed out by Deutscher (2008), the process discussed here essentially presupposes the existence of recursion, namely recursion within the noun phrase.

 Bernd Heine

2. Patterns of clause expansion Take the following example. In the Nigerian language Kanuri, the dative case enclitic -rò (dat), clearly an exponent of noun phrase syntax, can be attached to finite clauses like (1a) to form complement clauses (1b). (1) Kanuri (Saharan, Nilo-Saharan; Heine 1990) a. Sávā- nyī íshìn. friend- my come(3.sg) ‘My friend is coming.’ b. Sávā- nyī íshìn- rò tә̀mǎŋә́nà. friend- my come(3.sg)- dat thought.1.sg.perf ‘I thought my friend would come.’ I will say that (1b) is an instance of clause expansion, that is, of a conceptual strategy whereby clausal (propositional) participants are treated like nominal participants, and that this strategy has the effect that — over time — nominal structures acquire the properties of subordinate clauses (Givón 2008, ch. 5; Heine & Kuteva 2007). Even when this process has reached a more advanced stage, there tend to be some nominal properties that survive as relics, such as the following (but see also below):

(2) Structural properties commonly found on subordinate clauses arising via expansion

a. The marker of subordination resembles a grammatical form associated with noun phrase structure, such as a marker of case, gender, definiteness, or an adposition. b. The verb of the subordinate clause is non-finite, coded like an infinitival, gerundival, participial, or a nominalized constituent and takes the case marking of a corresponding nominal participant. c. The arguments of the subordinate clause are coded in a form that tends to differ from that of the main clause. d. The agent or notional subject takes a genitive/possessive or other case form, typically having the appearance of a genitival modifier of the subordinate verb. e. The patient or notional object may also take a genitive/possessive or other case form. f. There are severe restrictions on distinctions such as tense, aspect, modality, negation, etc. that can be expressed — in fact, such distinctions tend to be absent altogether.

From nominal to clausal morphosyntax 

The properties in (2) are not definitional ones; rather, they are taken to be diagnostic for identifying instances of expansion and, as we will see below, not all of the properties are necessarily present in a given case. To be sure, nominal encodings such as the ones listed in (2) are in no way restricted to specific languages; they are found in some way in quite a number of languages. For example, English He witnessed the enemy’s destruction of the city largely corresponds to (2), being a nominal version of the largely equivalent sentence He witnessed that/how the enemy destroyed the city. With reference to the four parameters of grammaticalization proposed by Heine and Kuteva (2007, section 1.2), clause expansion tends to have the following effects in particular: Extension means that an existing morphological device is extended from nominal to clausal structures, with the result that a new function, that of presenting subordinate clauses, emerges. This has the effect that the nominal function associated with this device is lost in the relevant contexts (desemanticization), and also that the ability associated with nominal structures to take determiners and modifiers is lost (decategorialization). Finally, erosion, which may but need not be involved, means that the marker of subordination tends to lose in phonetic substance, becoming shorter or phonetically simplified vis-à-vis the corresponding nominal marker. Which kinds of construction undergo expansion is determined, first, by the kind of subordination that is the target of expansion. For example, an overview of the relevant literature suggests that clause expansion is more likely to be observed in the development of complement clauses than in relative or adverbial clauses. Second, it is also lexically determined, in that it tends to affect some verbs more than others, most of all speech-act, cognition, volitional, and phase verbs, which typically take both nominal and propositional complements, such verbs being e.g., ‘see’, ‘hear’, ‘feel’, ‘want’, ‘finish’, ‘start’, ‘know’, ‘tell’, ‘remember’, ‘say’, etc. Frajzyngier (1996: 234) distinguishes in Chadic languages between ‘like’-verbs and ‘want’-verbs and concludes that the former tend to be associated with nominal complements while the latter imply a subsequent action or event and are more likely to take propositional complements. And, third, it is also determined by the structure of the language involved (see Givón 2008, ch. 5).

3. A five-stage scenario In order to reconstruct how new forms of clause subordination may arise via clause expansion, I carried out a crosslinguistic survey. The goal of the survey was to reconstruct the mechanism that can be hypothesized to be at work in the development from nominal to clausal complement morphosyntax. The sample employed was dictated by the availability of data; while it contains languages from a range of

 Bernd Heine

genetically and areally unrelated languages, no claim is made on whether it is in any way representative of the world’s languages at large. Complement clauses arising via expansion tend to be restricted to a limited spectrum of main clause (matrix) verbs (see section 2).

3.1 Nominal vs. verbal properties As a basis of reconstruction, a distinction between noun phrase and clausal morphosyntax is made. The former is said to manifest itself in the presence of what will be called “nominal properties” such as the ones listed in (3). (3) Nominal properties Na non-finite marking (nominalizing, infinitival, gerundival, participial, etc. morphology) Nb possessive modifiers Nc case affixes or adpositions Nd noun phrase word order Ne raising Nf other means (markers of definiteness or indefiniteness, nominal number markings, etc.) Clausal morphosyntax is described in terms of what I loosely refer to as “verbal properties”, in particular the ones listed in (4). (4)

Verbal properties Va personal verbal affixes or clitics Vb tense-aspect markers Vc agreement between verb and subject Vd clausal word order Ve clausal participant marking Vf other properties (verbal derivation, negation, etc.)

A few notes on some of these properties and the way they are treated in this paper seem in order. Non-finite forms (Na) on verbs typically consist of a “nominalizing” morpheme, and/or a case affix or adposition, but they do not normally take any other morphological elements. Nevertheless, there are languages where they also mark categories such as transitivity, tense, aspect, cf. the tense-aspect distinctions used with infinitives in English, Russian, Classical Greek, etc. (Noonan 1985: 58–9); for a particularly complex kind of nominalization marker, see Clendon (1988) on the Manjiljarra dialect of the Australian Western Desert language. Nevertheless, if there are grammatical categories on the non-finite verb that are typically associated with verbal morphosyntax then these are likely to show severe restrictions compared to the verbal morphosyntax of the main clause.

From nominal to clausal morphosyntax 

Property Nb means that the complement subject and/or object are coded typically, though not necessarily, as a possessive modifier of the complement verb. In one language, the West African Niger-Congo language Koromfe, we found a compounding construction instead of a possessive construction; thus, in the following example, the complement object appears as the first component of an endocentric compound (‘knife giving’): (5)

Koromfe (Gur, Niger-Congo; Rennison 1996: 44) a gabrε paʊ a kεɔ a kεkʊ joro kaŋәnaa. art knife give.nomin art woman art field in be.hard.prog ‘It’s hard to give a woman a knife in a field.’ (Lit.: ‘Knife giving a woman in the field is hard’)

While nominalization is a paradigm instance of noun phrase morphology, there are a number of languages that have no nominalizing morphology and in such cases I relied on other structural features to establish the presence of a nominal structure, in particular word order (Nd). For example, the Northern Khoisan language !Xun has hardly any affixal inflections or derivations; that it uses a nominal structure in complement clauses after certain verbs, such as kàlè ‘want’, can be concluded from the word order employed: This language has invariably verb-medial (SVO) order, cf. (6a), but in such complement clauses the order is OV, that is, the word order is that of attributive possession (6b). Thus, the sentence in (6b) can be translated literally as ‘I want the woman’s giving of water’, where the complement recipient ‘woman’ is coded as a possessive modifier of the complement verb, acting like a head noun in a possessive construction, while the complement theme (or patient) is presented by means of the transitive preposition kē (tr) via clausal participant marking. (6) !Xun (Northern Khoisan, W2 dialect; own field notes) a. mí má |àˈā dȁhmà kē gǁú. 1.sg top give woman tr water ‘I give the woman water.’ b. mí má kàlè dȁhmà |àˈā kē gǁú. 1.sg top want woman give tr water ‘I want to give the woman water.’

3.2 A scenario On the basis of differences in the treatment of these properties, a five-stage scenario is proposed for the process leading to the rise of one specific type of subordinate clauses — a process that is described by Givón (in this volume) as one where “the complement-clause event is treated analogically as a nominal object of the main clause.”

 Bernd Heine

3.2.1 The noun stage I hypothesize that at the beginning of the process leading to the type of complement clause subordination looked at in this paper there is a nominal complement or adjunct as, e.g., in English I want candies, I know that person.

3.2.2 The extended noun stage As observed above, our concern is with verbs that may have either a nominal or a propositional complement, and stage I relates exclusively to the latter. This stage concerns predications of what Noonan (1985: 60) calls nominalized complements with the internal structure of noun phrases. This is crosslinguistically a fairly common construction; in Haspelmath’s (2005 et al.: 502) world-wide sample of ‘want’ complement clauses of 283 languages, more than half (144) belong to this type. The main properties of this stage are: (7) Properties of stage I a. The complement or adjunct (C) is a non-finite verb (NFV), typically in a nominalized, infinitival, or participial form. b. The subject, to use Haspelmath’s (2005: 502) phrasing, tends to be “left implicit” in object complement clauses; it is coreferential with the matrix subject. c. The complement can be interpreted alternatively as a nominal constituent or a subordinate clause. d. Arguments of the NFV are encoded as oblique participants, typically as genitival modifiers, occasionally also as peripheral participants of the NFV. e. The complement subject or object of C may be coded as the object of the matrix clause (“raising”). f. The complement lacks most or all tense-aspect markings and other trappings characteristic of matrix clause verbs. g. Linear ordering is that of nominal rather than of verbal constituents. A paradigm instance of stage I is provided by the following example from English, where both the complement subject and object are presented as possessive modifiers: Algernon’s shooting of the aardvark drew international attention (Noonan 1985: 60). The following example from Estonian illustrates one of the two ways in which complement clauses having speech-act or mental-state verbs as main verbs are expressed in this language: The verb is non-finite, constructed in the present tense of the active participle, and the subject/agent appears in the genitive case (gen): (8)

Estonian (Finno-Ugric; Harris & Campbell 1995: 99) sai kuul- da seal ühe mehe ela- vat. got hear- inf there one.gen man.gen live- pres.active.ptcpl ‘S/he came to hear that a man lives there.’

From nominal to clausal morphosyntax 

In a number of languages there is no special morphology on the complement verb, that is, there is no morphological distinction between finite and non-finite verb forms; nevertheless, there may be other means which provide clues that we are dealing with nominal clauses. Such clues may consist of markers of attributive possession (Nb). For example, in the Chadic language Angas, nominalization is not marked, but the fact that the object is coded as a possessive modifier of the verb shows that there is nominalization (Frajzyngier 1996: 243): (9)

Angas (Chadic, Afroasiatic; Frajzyngier 1996: 243) Musa rot dyip kә̀- shwe. Musa want harvest poss- corn ‘Musa wants to harvest corn.’ (lit.: ‘Musa wants the harvesting of the corn.’)

Alternatively, it can be word order characteristics (Nd) that suggest that we are dealing with a nominal structure, as in our !Xun example of (6). In the West African language Hausa, the case-marking morphology appears on the complement verb rather than its nominal complement: It consists of the enclitic genitive linker (linker) -n, diachronically the masculine genitive marker (cf. (10a)), which connects the preceding complement verb, behaving like a head noun, with the following complement noun, being a possessive modifier. This possessive structure is used for both complement objects (10b) and complement subjectss (10c): (10) Hausa (Chadic, Afroasiatic; Newman 2000: 310, 311) a. bāya- n gàrī ‘the back of the town (or behind the town)’ back- m.linker town b. sun dainà shâ- n giyā̀. they quit drinking- linker beer ‘They quit drinking beer.’ c. har̃bì- n wàzīrì yā bur̃gē ̀ ni. shooting- linker vizier 3.sg.m impress me ‘The vizier’s shooting impressed me.’ Raising is considered here a nominal property even in languages where it has the status of an affix in the matrix clause, as in the following example: (11)

Bole (Chadic, Afroasiatic; Frajzyngier 1996: 263) ita ndol- na te- yyi. 3.f want- 1.sg eat- nomin ‘She wants me to eat.’

The following examples from Ancient Greek and Latin are also taken to be instances of stage I since the dative case (dat) of the complement clause presents an argument

 Bernd Heine

of the matrix verbs éksestin ‘it is possible’ of Greek and licet ‘it is permitted’ of Latin, respectively: (12) a.

Ancient Greek (Comrie 1997: 43) Nûn soi éksestin andrí genésthai. now you.dat it.is.possible man.dat to.become ‘Now is it possible for you to be a man?’

b.

Latin (Comrie 1997: 43) Mihi neglegenti esse non licet. I.dat negligent.dat to.be not it.is.permitted ‘It is not permitted for me to be negligent.’

Being an argument of the matrix clause, the NFV may have a case affix or adposition on it. But depending on the language concerned, it may as well be marked for other categories. Thus, there may be tense-aspect distinctions used with the NFV (see above). I am ignoring here adverbial adjuncts, which generally appear to be coded as clausal participants; an issue that I will return to below (section 4).

3.2.2.1 Evidence for transfer from nominal to verbal structure That there is in fact an extension from nominal to clausal morphosyntax may be illustrated with the following example from the Nilo-Saharan language Ik of Uganda. In the case system of this language there is one peculiarity: The main clause object appears in the accusative case (acc) whenever the subject has third person reference, cf. (13b), but in the nominative (nom) when the subject has first or second person reference (13a). The same case marking is found in object complement clauses, cf. (13c) and (13d). (13) Ik (Kuliak, Nilo-Saharan; König 2002) a. bεɗ- Ι á mes- a. want- 1.sg beer- nom ‘I want beer.’ b. bεɗ- a mes- íka. want- 3.sg beer- acc ‘He wants beer.’ c. bεɗ- Ι á atsˈ- ésa ŋƙáƙá- é. want- 1.sg eat- inf.nom food- gen ‘I want to eat food (or meat).’ (Lit.: ‘I want the eating of food’.) d. bεɗ- a atsˈ- és- íka ŋƙáƙá- é. want- 3.sg eat- inf- acc food/meat- gen ‘He wants to eat meat.’

From nominal to clausal morphosyntax 

The structure of the Ik complement clause is a canonical instance of stage I: The (non-finite) complement verb ‘to eat’ in (13c) appears in a non-finite form and is casemarked, and the object of the complement clause is treated like a possessive modifier in the genitive case (gen). Thus, complement clauses are structured on the model of nouns. As the discussion in Deutscher (2008) suggests, the transition from stage 0 to stage I appears to involve a process that is more complex that has hitherto been assumed; more research is required on this issue.

3.2.3 Stage II: Mixed morphosyntax The nominal structure is gradually intruded by a clausal syntax. At this stage, the complement clause is still determined by nominal structures but there are now elements of clausal morphosyntax that are also found in finite clauses, such as the ones listed in (14). (14) Properties of stage II a. One or more arguments are presented as clausal participants. This applies in particular to the complement object. b. Parts of the complement syntax are determined by the word order of finite clauses. c. The complement verb may have elements of finite verb morphology on it. Rather than coding the complement subject or object as a nominal modifier, the non-finite complement verb takes an object in much the same way as finite clauses do — that is, the complement is characterized by the presence of a [verb-object] constituent, as in the East African language Swahili, where (15a) is a main clause and (15b) an object complement clause: (15) Swahili (Bantu, Niger-Congo) a. Ali a- li- m- saidia Hadija. Ali n1.s- past- n1.o- help Hadija. ‘Ali (had) intended to help Hadija.’ b. Ali a- li- kusudia ku- m- saidia Hadija. Ali n1.s- past- intend inf- n1.o- help Hadija. ‘Ali (had) intended to help Hadija.’ The structure of the complement clause presents a mixture of nominal and clausal structures. Thus, in the English subject complement clause construction illustrated

 Bernd Heine

below, the subject (Cartier) has a nominal structure while the object (Dugué) is coded like a main clause object. (16) English Cartier’s defeating Dugué is significant.

(Noonan 1985: 43)

In a similar fashion, in the following example from Uzbek, the complement subject is coded in the genitive like a possessive modifier while the object shows clausal syntax, taking the object case marking. Note that there is an inflected complement verb, but the suffix -i- is not one of main clause syntax but rather it “reinforces the associative relationship” (Noonan 1985: 61). (17) Uzbek (Noonan 1985: 60) Xɔtin bu ɔdam- niŋ j ̌oj ̌a- ni oǧirla- š- i woman this man- gen chicken- obj steal- nomin- 3.sg ni istadi. obj wanted.3.sg ‘The woman wanted the man to steal the chicken.’ In the following example from the Tungusic language Evenki, the complementizer is an accusative case marker (acc), that is, the complement clause is introduced by a case suffix, in accordance with Nc in (3), the verb әmә- ‘come’ of the complement clause is presented in the resultative participle (part), cf. Na in (3), and the agent of the complement clause appears as a possessor suffix (-s ‘your’) on the participle form of the verb. But in addition to these nominal structures there are also clausal ones, such as the subject pronoun si: (18) Evenki (Tungusic; Comrie 1981: 83) әnii- m әә- ćәә- n saa- rә si tәnәwә mother- my neg- past- 3.sg know- ?4 you yesterday әmә- nәә- wәә- s. come- part- acc- 2.sg ‘My mother doesn’t know that you arrived yesterday.’ Another typical mixed situation can be illustrated with the following example from the Krongo language of the Kordofan Hills of Sudan. There are both nominal and verbal properties on the verb of the object complement clause: The nominalization and the second person possessive markers are suggestive of the former, and the verbal

4. No glosses are provided by the author.

From nominal to clausal morphosyntax 

derivation (BEN) and transitivity markers (TR) of the latter. Furthermore, the direct object (ɲàamà) and the beneficiary (àɁàŋ) also appear to be coded as clausal participants, cf. (19a). The same kind of mixed situation is found in the second type of object complement clause of Krongo, which involves subject-to-object raising (ʊ̀Ɂʊ̀ŋ), cf. (19b). (19) Krongo (Kordofanian; Reh 1985: 333–7) a. n- átàasà àɁàŋ t- óshó- ókò- n- tú 1/2- want I nomin- impfv.cook- ben- tr- 2.sg

ɲàamà àɁàŋ. things dat..I

‘I want you to cook for me.’ (Lit.: ‘I want your cooking for me.’)

b. n- átàasà àɁàŋ ʊ̀Ɂʊ̀ŋ kú- t- úmúnó àɁàŋ. 1/2- want I you loc- nomin- impfv.help me ‘I want you to help me.’ (19b) illustrates a common stage II situation where the complement verb shows nominal properties whereas its participants are all characterized by verbal (clausal) codings. Similarly, the following complement clause type of the Ethiopian language Maale marks the complement verb in the infinitive (plus appropriate case suffix) while all of its participants (except the complement subject) are presented like main clause participants: (20)

Maale (Omotic, Afroasiatic; Amha 2001: 177) Ɂála Ɂú∫k- itsí nayí- m k’ára t- uwá- se. beer.abs drink- inf.nom child.abs- dat good be- ipfv.neg- neg ‘Drinking beer is not good for a child.’

3.2.4 Stage III: Clausal syntax with nominal relics The complement is now a full-fledged subordinate clause. Still, there are relics of nominal morphosyntax that bear witness to its nominal origin. The clearest case is provided by languages where the morphosyntax of the subordinate clause is largely or entirely identical to that of main clauses and the only relic is a case marker or some other element of nominal morphology. Thus, we saw in section 2 that in Kanuri, the dative case enclitic -ro is found on complement clauses, which otherwise have the structure of finite main clauses, and it appears to be the only relic of the erstwhile nominal structure, otherwise complement and adverbial clauses behave like other finite clauses (Noonan 1985: 47; Heine 1990). And in Imbabura Quechua it is the accusative case marker (acc)

 Bernd Heine

in particular that bears witness to the nominal origin of the complement clause, which is finite: (21)

Imbabura Quechua (Cole 1982: 43) Pedro ya- n [ñuka Agatu- pi kawsa- ni] -ta. Pedro think- 3 I Agato- in live- 1- -acc ‘Pedro thinks that I live in Agato.’

In the Caucasian language Laz of Turkey it is possible to have a dative marker cliticized to a finite verb form, thereby turning a main clause, as in (22a), into a subordinate one (22b): (22) Laz (South Caucasian; Nino Amiridze; Funknet, April 2005) a. ali oxori- sha mo- xt- u. Ali house- in preverb- come- s3.sg.aor ‘Ali came home.’ b. ali oxori- sha mo- xt- u- shi […]. Ali house- in preverb- come- s.3.sg.aor- dat ‘When Ali came home […].’ Similarly, in the Ethiopian language Maale, a nominalized complement clause (23b) can be distinguished from a main clause (23a) only by the fact that it takes the nominalization marker -tsí instead of a declarative marker (DCL): (23) Maale (Omotic, Afroasiatic; Amha 2001: 177) a. nu Ɂá∫ínna- á jink- ó Ɂááɖ- á- ne. 1.pl.gen neighbors- nom Jinka- abs go- ipfv- dcl ‘Our neighbours are going to Jinka.’ b. nu Ɂá∫ínna- á jink- ó Ɂááɖ- á- tsí 1.PL.gen neighbors- nom Jinka- abs go- ipfv- nomin

goné- ke. true- be.dcl

‘It is true that our neighbors are going to Jinka.’

In the Squamish language of British Columbia, all nominals are accompanied by an article (art), and so are nominalized complements, as the description by Noonan (1985: 61) suggests. Complement clauses such as the following have all of the verbal inflections, clitics, and sentence particles to be found in main clauses; still, the presence of an article in the complement clause bears witness to the nominal origin of the structure.

From nominal to clausal morphosyntax 

(24) Squamish (Noonan 1985: 61) č- n ɬč- iws kwi n- s- na wa decl- 1.sg tired- body art 1.sg.poss- nom- fact prog cˈaqˈ- an- umi. hit- trans- 2.sg.obj ‘I’m tired of hitting you.’

3.2.5 Stage IV: The full-fledged complement clause Finally, there are complement clauses that are indistinguishable in their morphosyntax from finite main clauses, as in the following example, where the object complement clause (25a) is structurally identical with the main clause (25b) except for the topic marker má (top), which is mandatory in declarative main clauses but may not be used in subordinate clauses: (25) !Xun (North Khoisan, Khoisan; own field notes) a. mí m- é bhȁlì mí dàbà ǁˈàn. 1.sg top- past dream 1.sg child be.sick ‘I was dreaming that my child is sick.’ b. mí dàbà má ǁˈàn. 1.sg child top be.sick ‘My child is sick.’ Stage IV situations may be due to two different processes: (a) Either there was an evolution such as the one sketched above, with the result that all nominal properties have disappeared, or (b) there never was a nominal construction; rather, the structure of the main clause is copied in toto onto the subordinate clause. While (a) is suggestive of clause expansion, (b) is an instance of clause integration, where two distinct clauses are combined into one complex sentence (cf. Givón’s clause-chaining pathway; see section 1). Which of the two, (a) or (b), is involved is hard to determine in many cases.

4. Some generalizations The extent to which nominal and verbal properties contribute to structuring complement clauses is shown in table 1 on the basis of the fairly small sample that is used in this study (see Appendices 1, 2). As the figures in table 1 suggest, it is the complement predicate that stands out as showing the highest amount of nominal properties (78.9 %), followed by the subject (69.9 %) and the object (21.4 %). An extreme situation is found with “other participants”, which almost invariably are adjuncts: They are associated exclusively with verbal properties. Assuming that these figures are suggestive of a diachronic process from nominal to clausal morphosyntax, one may hypothesize that the process starts out with peripheral

 Bernd Heine

Table 1. Relative contribution of nominal vs. verbal properties in structuring complement clauses showing nominal properties (in percentages. N = nominal properties, V = verbal properties, O.Com = object complement clause, S.Com = subject complement clause). Predicate Type of clause

Subject

Object

Other participants

N

V

N

V

N

V

N

V

O.Com (18)

81.0

19.0

58.3

41.7

30.0

70.0

0

100

S.Com (13)

76.5

23.5

81.1

18.9

0

100.0

0

100

Total (31)

78.9

21.1

69.9

30.1

21.4

78.6

0

100

participants (adjuncts), subsequently affecting complement objects, subjects, and finally the complement predicate, as sketched in the following scale: (26) predicate < subject < object < adjunct Note that this scale is probabilistic in nature: It predicts what is likely to happen rather than what must happen. What the scale captures is the following: Adjuncts (peripheral participants) of complement clauses are the first to be coded by means of verbal morphosyntax; in fact, they are likely to appear already at stage I as clausal participants indistinguishable from main clause adjuncts. The next to acquire the properties of clausal syntax are (direct) objects. This observation also surfaces in Noonan’s (1985: 61) analysis: He observes that cases such as Irish, where only the complement (“notional”) object shows a possessive syntax (an “associative relation”), that is, a nominal property, are rare. Compared to other complement participants, complement subjects appear to be the most resistant to change; but clearly the most conservative of all is the predicate structure, which tends to retain nominal (or nominalizing) properties when other constituents of the clause have lost them. The scale in (26) can be read on the one hand as a synchronic implicational structure of the kind “If any of the categories of the scale is characterized by a nominal property then all categories to its right are also likely to be”. On the other hand, I hypothesize that the scale can also be interpreted as a diachronic scenario, describing the growth of complement clauses out of nominal complements via clause expansion or, in more general terms, a grammaticalization process leading from nominal to clausal morphosyntax.

5. Evidence for the development from nominal to propositional structures That there is a fairly common grammaticalization process leading from nominal to clausal morphosyntax can be shown by looking at other kinds of grammaticalization;

From nominal to clausal morphosyntax 

the development proposed in the preceding sections is but one manifestation of this strategy. In fact, there is some evidence suggesting that conceptualizing and describing propositional contents, typically expressed by clauses, in terms of concrete objects, coded linguistically as nouns, is a salient human strategy. First, nominalization of subordinate clauses is not restricted to complement clauses; it also concerns relative and adverbial clauses, as aptly demonstrated by Givón (1994; in this volume), who observes for example: In many language families — Turkic, Carribean, Bodic (Tibeto-Burman), No. Uto-Aztecan, Sumerian, to cite only a few — all subordinate clauses are nominalized, at least historically. Such structures may re-acquire finite properties over time (Givón 1994; Watters 1998), but the morphology retains, for a long time, the telltale marks — clear fossil evidence — of the earlier nominalized status. (Givón in this volume)

Second, there are some well documented grammaticalization processes whose main effect is that noun phrase morphology is extended to introduce clauses. Thus, demonstrative attributes on nouns commonly grammaticalize to relative clause markers, and nominal case markers turn into markers of clause subordination. Another piece of evidence can be seen in the following observation made by Deutscher (2008), according to which “the morphemes that we see synchronically in the role of nominalizing verbs all somehow appear to go back to elements that were originally attached to nouns”, that is, the use of nominal elements is extended to verbs. Third, in the rise of new tense and aspect morphologies it may happen that participant roles reserved for nominal constituents are extended to take clausal/propositional constituents; thus, structures such as (27a) commonly develop diachronically into structures like (27b) (see Heine & Kuteva 2002 for more examples): (27) English a. He used all the money. b. He used to visit her once a month. Fourth, there is also a well documented lexical process whereby negative existential verbs taking a nominal participant (‘there is no X’) may be extended to take clausal participants (‘there is no doing of X’), thereby giving rise to clausal negation markers. For example, in Mandarin, the negative existential méi [yŏu] takes nominal complements, as in (28a), but its use appears to have been extended to verbal complements, as in (28b), with the result that there now is a new negation marker of completed actions (for more examples and details of this process, see Croft 1991): (28) Mandarin (Croft 1991: 11; cited from Li & Thompson 1981) a. méi [yŏu] rén zài wìmian. neg.exist person at outside ‘There’s no one outside.’

 Bernd Heine

b. ta- méi [you] siˇ. 3.sg neg.exist die ‘S/he hasn’t died’, or ‘She didn’t die.’ Finally, that there is a unidirectional development whereby the use of nominal structures is extended to verbal structures can also be demonstrated with the following example. A typological survey of question pronouns suggests that there is a widespread process whereby interrogative pronouns referring to inanimate objects (‘what?’) are extended to also refer to actions and events (Heine, Claudi & Hünnemeyer 1991: 56ff.). Evidence for this directionality comes in particular from languages where the interrogative pronoun is etymologically transparent: In such languages the pronoun is not infrequently derived from a phrase ‘which thing?’ For example, in the Ewe language of Togo and Ghana, the pronoun nú-ka ‘what?’ means historically ‘thing-which?’, but is used in the same way for nominal as for verbal referents, as in (29), and the interrogative pronoun m̄tcí ‘what?’ of the !Xun language of southwestern Africa, which is historically composed of the interrogative element *m̄5 and the noun tcí ‘thing’, is not restricted to nominal referents but is used in much the same way also to refer to actions and events, cf. (30). (29)

Standard Ewe (Kwa; Niger-Congo; own data) nú-ka wɔ- ḿ ne- le? thing-which do prog 2.sg- prog ‘What are you doing?’

(30)

!Xun (W2 dialect, North Khoisan; own field notes) m̄tcí á hȁ- è o? Q.thing Q n1 rel do ‘What does he do?’

Further data are found in pidgins and creoles, where not uncommonly the question word referring to actions (‘what?’) is transparently derived from the phrase ‘which thing?’, as in the following example from the Spanish-based creole Papiamentu: (31)

Papiamentu (Holm 1988: 180) Ta kiko Wan ta hasi? is what.thing John TAM do ‘What is John doing?’

5. *m- is no longer a productive morpheme in !Xun.

From nominal to clausal morphosyntax 

6. Conclusions The “plasticity” of linguistic manipulation as it manifests itself in the development from less complex (nominal) structures to more complex (clausal) ones can be likened to the kind of neuroplasticity as it has been observed in the development of nervous systems (Tublitz, this volume). The ultimate motivation underlying the linguistic process, however, is largely unclear. To be sure, the “need” to perform complex novel functions can be invoked as possible driving force but, to the best of my knowledge, no reasonable homologue, or analogue, to the biological notion of adaptation has so far been identified. In a critical discussion of the present paper, Deutscher (this volume) observes that one can decide to define subordination in such a way that a sentence that we presented earlier, reprinted here as (32), would not be considered a subordinate structure, but that we cannot take (32) as our point of departure for analysis. I agree that the issue under discussion is — at least to some extent — one of definition. For the present purposes I have been assuming that a subordinate clause is typically a finite clause, less typically a (non-finite) clause with a reduced verbal morphosyntax, while a noun phrase — whatever its syntactic function may be — is not a subordinate clause. More importantly, as we saw in section 3, structures illustrated by (32) are not taken as a point of departure in my analysis: They are instances of stage I, that is, products of a (hypothesized) process leading from stage 0 to stage I. (32) English Algernon’s shooting of the aardvark drew international attention. One may argue about which of the stages distinguished in section 3 are the most interesting for understanding the growth of clausal out of nominal structures. Deutscher finds only the process from stage 0 to stage I to be “cognitively challenging”,6 but I do not think that it is the only one that is really challenging. An equally challenging question for me is whether the gradual loss of nominal properties in favor of verbal properties, that is, the process leading from stage I to stage IV, has any extra-linguistic cognitive correlates, and if yes, what these correlates are — an issue that requires much further research. I have avoided the term “complexity” in this paper and how it is to be defined. In terms of the threefold distinction proposed by Bowern (2008) one could argue that our concern was exclusively with ontological complexity: To the extent

6. “[…] the only step in the process that is cognitively challenging, the pressure-packaging of a verb as a noun in the first place.” (Deutscher 2008: 3).

 Bernd Heine

that the inherent nature of a finite clause can be argued to be morphosyntactically more “complex” than a noun, we were dealing with a process from less to more complex structure. The case I discussed in this paper relates to both Dahl’s (2004) system complexity and structural complexity: The process hypothesized leads from a less to a more complex type of complement, and from lower to higher degree of structural complexity of expression (see also Dahl 2008: 153–4), and appears to be one of gradual change towards increased complexity, the end product being a complement clause whose structure is largely indistinguishable from that of finite main clauses. The hypothesis proposed in the present paper is far from new. That new forms of complex sentences arise via clause integration has been argued for by a number of authors (see especially Hopper & Traugott 2003; Givón 2005; 2006; 2007; Heine & Kuteva 2007), even if not accepted by all (see Deutscher, this volume). The objective of this paper was a narrow one. First, I was restricted to complement clauses and, second, my concern was exclusively with clause expansion. But even the rise of complex sentences via the expansion of simple sentences has already been dealt with in some way or other in earlier works (e.g., Givón 1994; 2002; 2005; 2008; in this volume; Heine & Kuteva 2007); the question that I was concerned with here was with the nature of one kind of process leading from nominal to clausal structures. As I argued in section 3, there are a number of stages of development leading from fully nominal complements at stage 0 to fully clausal constructions at stage IV, with each new stage characterized by a decrease in the amount nominal properties and an increase in verbal and clausal properties. In section 4 we saw that that this gradual process appears to have an internal structure of the following kind: It affects first adjuncts, which are coded as clausal participants from stage I on, followed by clausal objects, which again tend to be followed by subjects, and it is the verbal morphosyntax that turns out to be the most conservative component of complement constructions, surviving as a rule up until stage III. At the final stage IV there are no more traces of nominal morphosyntax — the complement clause is now largely or entirely identical with the main clause. One may speculate that the similarities shared by the implicational scale presented in (26) and other scales that have been devised ever since Silverstein (1976) and Keenan and Comrie (1977) proposed their feature or accessibility hierarchies7 are not coincidental. Note further that a hierarchy of this kind can also be observed in first language acquisition; as Diessel (this volume) shows, “subject relatives (S- and A-relatives) caused fewer errors than direct object relatives (P-relatives), which in turn caused fewer errors than indirect object relatives (io-relatives) and adverbial relatives (adv-relatives)” among English- and German-speaking children.

7. Cf. also Langacker’s (1997: 262) reference-point chain subject > object > other.

From nominal to clausal morphosyntax 

This, however, is an issue that would require a separate analysis. What the present survey shows is that neither the scenario of section 3 nor the scale in (26) correlates significantly with languages as a whole but rather with specific constructions of a given language. Quite commonly there are two or more complement clause constructions within one and the same language, where each of the constructions represents a different stage of development or, even more commonly, where one construction is suggestive of clause integration and the other of clause expansion, as in the following Finnish example: Whereas (33a) can be assumed to be an instance of integration, (33b) appears to represent clause expansion of stage II, where there are both nominal properties (cf. the coding of the complement subject as a genitival modifier) and clausal properties (the locative argument Helsingissä is coded like a main clause participant). (33) Finnish (Comrie 1997: 45) a. Tiedän, että sinä olet Helsingissä. I.know that you.nom are in.Helsinki

or

b. Tiedän sinun olevan Helsingissä. I.know you.gen being in.Helsinki ‘I know that you are in Helsinki.’ It would be interesting to relate the stages distinguished to possible differences in the prosodic structures characterizing clause combining (see Mithun, this volume); but on the basis of the present data, taken overwhelmingly from published sources, this is not possible. The hypotheses presented were based on findings made in studies on grammaticalization. For example, case affixes and adpositions have been shown to commonly develop into markers of clause subordination while a development in the opposite direction is unlikely to happen; hence the conclusion drawn in this paper is that if there is a morphological element in a given language that serves both as a case marker or adposition and as an element introducing complement clauses then the former is the older function. Grammaticalization rests on generalizations on grammatical change, that is, it is diachronic in nature and, accordingly, relies on and can be falsified by means of historical evidence. But so far not much historical evidence has become available on the reconstructions proposed in this paper; accordingly, the conclusions reached here have to be taken with care until such evidence becomes available.

Abbreviations abs = absolutive; art = article; assoc = associative; c = complement; caus = causative; comp = complementizer; cop = copula; dat = dative; dcl, decl = declarative; gen = genitive; inf = infinitive; ipfv, impfv = imperfective; loc = locative; n = nominal

 Bernd Heine

property; n1 = noun class 1 or third person singular pronoun; neg = negative; nfv = non-finite verb; nom = nominative; nomin = nominalizer; obj = object; perf = perfect; pl = plural; poss = possessive; prog = progressive; sg = singular; top = topic marker; tr = transitivity marker; v = verbal property; vn = verbal noun; 1, 2, 3 = first, second, third person.

Appendix 1. Data on nominal properties in complement clauses Angas (Chadic, Afroasiatic; Frajzyngier 1996: 243) Musa rot dyip kә̀- shwe. Musa want harvest poss- corn ‘Musa wants to harvest corn.’ (lit.: ‘Musa wants harvesting of the corn.’) Bole (Chadic, Afroasiatic; Frajzyngier 1996: 263)

ita ndol- na te- yyi. 3.f want- 1.sg eat- nomin ‘She wants me to eat.’

English a. b. c.

Burt’s being a chicken farmer worries Max. (Noonan 1985: 49) Cartier’s defeating Dugué is significant. (Noonan 1985: 43) For Cartier to defeat Dugué would be significant. (Noonan 1985: 43) I want her to come.

Estonian (Finno-Ugric; Harris & Campbell 1995: 99)

sai kuul- da seal ühe mehe ela- vat. got hear- inf there one.gen man.gen live- pres.active.ptcpl ‘S/he came to hear that a man lives there.’

Evenki (Tungusic; Comrie 1981: 83)

әnii- m әә- ćәә- n saa- rә si tәnәwә mother- my neg- past- 3.sg know- ?8 you yesterday

әmә- nәә- wәә- s. come- part- acc- 2.sg

‘My mother doesn’t know that you arrived yesterday.’

8. No glosses are provided by the author.

From nominal to clausal morphosyntax 

Finnish (Comrie 1997: 45)

Tiedän sinun olevan Helsingissä. I.know you.gen being in.Helsinki ‘I know that you are in Helsinki.’

Ancient Greek (Comrie 1997: 43)

Nûn soi éksestin andrí genésthai. now you.dat it.is.possible man.dat to.become ‘Now is it possible for you to be a man?’

Hausa (Chadic, Afroasiatic; Newman 2000: 311–2) a. sun dainà shâ- n giyā̀. they quit drinking- linker beer ‘They quit drinking beer.’ b. har̃bì- n wàzīrì yā bur̃gē ̀ ni. shooting- linker vizier 3.sg.m impress me ‘The vizier’s shooting impressed me.’ Ik (Nilo-Saharan; König 2002) bεɗ- Ι ́a atsˈ- ésa

ŋƙáƙá- é. want- 1.sg eat- inf.nom food- gen ‘I want to eat food (or meat).’ (Lit.: ‘I want the eating of food’.)

Irish (Noonan 1985: 61)

Is ionadh liom Seán a bhualadh Thomáis. cop surprise with.me John comp hit.nomin Thomas.gen ‘I’m surprised that John hit Thomas.’

Kanuri (Saharan, Nilo-Saharan; Noonan 1985: 47)

Sávā- nyī íshìn- rò tә̀mǎŋә́nà. friend- my come(3.sg)- dat thought.1.sg.perf ‘I thought my friend would come.

Khwe (Central Khoisan, Khoisan)

xàcí tcà- á- tè ˈà tí |xˈân qāámà- à- tè. she be.sick- JUNC- pres acc I very regret- JUNC- pres ‘I am a lot sorry that she is sick.’

Koromfe (Gur, Niger-Congo; Rennison 1996: 44)

a gabrε paʊ a kεɔ a kεkʊ joro kaŋәnaa. art knife give.nomin art woman art field in be.hard.prog ‘It’s hard to give a woman a knife in a field.’ (Lit.: ‘Knife giving a woman in the field is hard’)

 Bernd Heine

Krongo (Kordofanian; Reh 1985: 333–337) a. n- átàasà àɁàŋ t- óshó- ókò- n- tú. 1/2- want I nomin- impfv.cook- ben- tr- 2.sg

ɲàamà àɁàŋ things dat.I

‘I want you to cook for me.’ (Lit.: ‘I want your cooking for me.’)

b. n- átàasà àɁàŋ ʊ̀Ɂʊ̀ŋ kú- t- úmúnó àɁàŋ. 1/2- want I you loc- nomin- impfv.help me ‘I want you to help me.’ Latin (Comrie 1997: 43)

Mihi neglegenti esse non licet. I.dat negligent.dat to.be not it.is.permitted ‘It is not permitted for me to be negligent.’

Maale (Omotic, Afroasiatic; Amha 2001: 173, 177) a. Ɂála Ɂú∫k- itsí nayí- m k’ára t- uwá- se. beer.abs drink- inf.nom child.abs- dat good be- ipfv.neg- neg ‘Drinking beer is not good for a child.’ b. Ɂízó- ko timirto máári Ɂááɖ- is’- á 3.f.sg.abs- gen school house.abs go- inf- nom

koɁ- is- á- ya- ke. want- caus- ipfv- nomin- be.dcl

‘Her going to school is necessary.’

c. nu Ɂá∫ínna- á jink- ó Ɂááɖ- á- tsí goné- ke. 1.pl.gen neighbor- nom Jinka- abs go- ipfv- nomin true- be.dcl ‘It is true that our neighbours are going to Jinka.’ Imbabura Quechua (Cole 1982: 43)

Pedro ya- n [ñuka Agatu- pi kawsa- ni] -ta. Pedro think- 3 I Agato- in live- 1- -acc ‘Pedro thinks that I live in Agato.’

Laz (South Caucasian; Nino Amiridze; Funknet, April 2005) a. ali oxori- sha mo- xt- u. Ali house- in preverb- come- s3.sg.aor ‘Ali came home.’

From nominal to clausal morphosyntax 

b. ali oxori- sha mo- xt- u- shi. Ali house- in preverb- come- s.3.sg.aor- dat ‘When Ali came home (…).’ Mandarin (Li & Thompson 1981: 575–81) a. zhòng shuǐguǒ de hěn nán guòhuó. grow fruit nomin very difficult make.living ‘It is difficult for fruit growers to make a living.’ b. nǐ méi yǒu wǒ xǐhuān de. you not exist I like nomin ‘You don’t have what I like.’ Manjiljarra (Clendon 1988: 195)

Mama- partarnu- nga nyangu mitu ngarri- nja- n. father- KIN- top saw dead lie- nomin- cont ‘He saw his father dead.’

Persian (Noonan 1985: 85)

Mæn adæd- æn- e Babæk- ra færman dadæm. I come- nomin- assoc Babak- obj order gave.2.sg ‘I ordered Babak to come.’

Imbabura Quechua (Cole 1982: 43)

Pedro ya- n [ñuka Agatu- pi kawsa- ni] -ta. Pedro think- 3 I Agato- in live- 1- -acc ‘Pedro thinks that I live in Agato.’

Squamish (Noonan 1985: 61)

č- n ɬč- iws kwi n- s- na wa decl- 1.sg tired- body art 1.sg.poss- nom- fact prog

cˈaqˈ- an- umi. hit- trans- 2.sg.obj

‘I’m tired of hitting you.’

Swahili

Ali a- li- kusudia ku- m- saidia Hadija. Ali n1.s- past- intend inf- n1.o- help Hadija. ‘Ali (had) intended to help Hadija.’

 Bernd Heine

Turkish (Kerslake 2007: 236–7) a. [Bura- ya kadar gel- me- miz] zor ol-uyor. here- dat as.far.as come- VN- 1.pl.poss difficult be-impf ‘It’s difficult for us to come all this way.’ b. Ali [bu araba- yi kullan- ma-] ya baʂla- dΙ. Ali this car- acc use- VN- dat begin- pl ‘Ali has begun to use this car.’ Uzbek (Noonan 1985: 60)

Xɔtin bu ɔdam- niŋ ǰoǰa- ni oǧirla- š- iwoman this man- gen chicken- obj steal- nomin- 3.sg-

ni istadi. obj wanted.3.sg

‘The woman wanted the man to steal the chicken.’

!Xun (Northern Khoisan, W2 dialect; field notes)

mí má kàlè dȁhmà |àˈā kē 1.sg top want woman give tr ‘I want to give the woman water.’

gǁú. water

Appendix 2. Nominal vs. verbal properties of complement clauses. (The symbols “Na”, “Vd”, etc. refer to the categories distinguished in (1) and (2). Verbal properties are printed in bold). Construction

Type of clause

Predicate

Subject

Object

Other participants

Angas

O.Com

Na

-

Nb

Bole

O.Com

Na

Va, Ve

English a

S.Com

Na

Nb

Ve

Ve

English b

S.Com

Na

Nc

Ve

Ve

English c

O.Com

Na

Ne

Ve

Ve

Estonian

S.Com

Na

Nb

Evenki

O.Com

Na

Nb, Va

Finnish

O.Com

Na

Nc

Ancient Greek

S.Com

Na

Nc

Hausa a

O.Com

Na

-

Hausa b

S.Com

Na

Nb

Ik

O.Com

Na

Nb

Irish

S.Com

Na

Nb

Ve Nb

From nominal to clausal morphosyntax 

Appendix 2. (Continued) Khwe

S.Com

Nc, Va, Vb, Vd

Ve

Ve

Kanuri

O.Com

Nc, Va, Vb

Ve

Ve

Krongo a

O.Com

Na, Vb, Vf

Nb

Ve

Krongo b

O.Com

Na, Vb, Vf

Ne

Ve

Latin

S.Com

Na

Nc

Laz

O.Com

Nc

Ve

Maale a

S.Com

Na, Vb

Maale b

S.Com

Na, Nc

Nc

Ve

Maale c

S.Com

Na, Vb

Ve

Ve

Mandarin a

S.Com

Na, Vd

Mandarin b

O.Com

Na, Va, Vd

Persian

O.Com

Na, Nb

I. Quechua

O.Com

Nc

Ve

Ve

Swahili

O.Com

Na

-

Turkish a

S.Com

Na

Nc

Turkish b

O.Com

Na

-

Ve

Uzbek !Xun

O.Com O.Com

Na -

Nb -

Ve Nd

Ve

Ve Ve

Va

Ve

Ve Ve

Ve

References Amha, A. 2001. The Maale Language. Leiden: University of Leiden. Austin, P. (Ed.). 1988. Complex Sentence Constructions in Australian Languages [Typological Studies in Language 15]. Amsterdam: John Benjamins. Bowern, C. 2008. Defining complexity: Historical reconstruction and Nyulnyulan. Paper presented at the International Symposium on the Rise of Syntactic Complexity, March 2008; Rice University. Bybee, J.L., Perkins, R.D. & Pagliuca, W. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago IL: University of Chicago Press. Bybee, J.L., Haiman, J. & Thompson, S.A. (Eds). 1997. Essays on Language Function and Language Type: Dedicated to T. Givón. Amsterdam: Jon Benjamins. Clendon, M. 1988. Some features of Manjiljarra nominalised relative clauses. In Austin 1988, 193–204. Cole, P. 1982. Imbabura Quechua. Amsterdam: North-Holland. Comrie, B. 1981. Language Universals and Linguistic Typology. Chicago IL: University of Chicago Press. Comrie, B. 1997. The typology of predicate marking. In Bybee et al. 1997, 39–50. Croft, W. 1991. The evolution of negation. Journal of Linguistics 27: 1–27.

 Bernd Heine Croft, W., Denning, K. & Kemmer, S. (Eds). 1990. Studies in Typology and Diachrony. Papers Presented to Joseph H. Greenberg on his 75th Birthday [Typological Studies in Language 20]. Amsterdam: Joh Benjamins. Dahl, Ö. 2004. The Growth and Maintenance of Linguistic Complexity [Studies in Language Companion Series 71]. Amsterdam: John Benjamins. Dahl, Ö. 2008. Two pathways of grammatical evolution. Paper presented at the International Symposium on the Rise of Syntactic Complexity, 27–29. 3. 2008; Rice University. Deutscher, G. 2008. Nominalization and the origin of subordination. Paper presented at the International Symposium on the Rise of Syntactic Complexity, 27–29. 3. 2008; Rice University. Diessel, H. 2004. The Acquisition of Complex Sentences [Cambridge Studies in Linguistics 105]. Cambridge: CUP. Dixon, R.M.W. (Ed.). 1976. Grammatical Categories in Australian Languages. Canberra: Australian Institute of Aboriginal Studies. Frajzyngier, Z. 1996. Grammaticalization of the Complex Sentence: A Case Study in Chadic [Studies in Language Companion Series 32]. Amsterdam: John Benjamins. Givón, T. 1979. From discourse to syntax: Grammar as a processing strategy. In Discourse and Syntax, T. Givón (Ed.), 81–109. New York NY: Academic Press. Givón, T. 1994. Nominalized clauses in Ute: The diachronic seesaw of finite and non-finite structure. 2ndo Encuentro de Linguística en el Noroeste. Hermosillo: Universidad de Sonora. Givón, T. 1995. Functionalism and Grammar. Amsterdam: John Benjamins. Givón, T. 2002. Bio-linguistics: The Santa Barbara Lectures. Amsterdam: John Benjamins. Givón, T. 2005. Context as Other Minds. Amsterdam: John Benjamins. Givón, T. (in this volume). Multiple routes to clause-union: The diachrony of syntactic complexity. Typescript, Seminario Sobre Complejidad Sintáctica, Hermosillo: Universidad de Sonora. Givón, T. 2008. The Genesis of Syntactic Complexity: Diachrony, Ontogeny, Neuro-Cognition, Evolution. Amsterdam: J. Benjamins. Harris, A.C. & Campbell, L. 1995. Historical Syntax in Cross-Linguistic Perspective. Cambridge: CUP. Haspelmath, M., Dryer, M.S., Gil, D. & Comrie, B. (Eds). 2005. The World Atlas of Language Structures. Oxford: OUP. Heine, B. 1990. The dative in Ik and Kanuri. In Croft et al. 1990, 129–149. Heine, Bernd & Mechthild Reh. 1984. Grammaticalization and reanalysis in African languages. Hamburg: Buske Heine, B., Claudi, U. & Hünnemeyer, F. 1991. Grammaticalization: A Conceptual Framework. Chicago IL: University of Chicago Press. Heine, B. & Kuteva, T. 2002. On the evolution of grammatical forms. In Wray 2002, 376–97. Heine, B. & Kuteva, T. 2007. The Genesis of Grammar: A Reconstruction. Oxford: OUP. Holm, J.A., 1988. Pidgins and Creoles, Vol. 1: Theory and Structure. Cambridge: CUP. Hopper, P.J. & Traugott, E.C. 2003. Grammaticalization. Cambridge: CUP. Keenan, E.L. & Comrie, B. 1977. Noun phrase accessibility and universal grammar. Linguistic Inquiry 8: 63–99. Kerslake, C. 2007. Alternative subordination strategies in Turkish. In Rehbein et al. 2007, 231–58. König, C. 2002. Kasus im Ik [Nilo-Saharan 17]. Cologne: Rüdiger Köppe. Langacker, R.W. 1997. A dynamic account of grammatical function. In Bybee et al. 1997, 249–73.

From nominal to clausal morphosyntax 

Lehmann, C. 1982. Thoughts on Grammaticalization. A Programmatic Sketch, Vol. 1. AKUP 48 [Arbeiten des Kölner Universalien-Projekts]. Cologne: Universität zu Köln, Institut für Sprachwissenschaft. Li, C.N. & Thompson, S.A. 1981. Mandarin Chinese. A Functional Reference Grammar. Berkeley LA: University of California Press. Mithun, M. 2008. Re(e)volving complexity: Adding intonation. Paper presented at the International Symposium on the Rise of Syntactic Complexity, 27–29. 3. 2008; Rice University. Newman, P. 2000. The Hausa Language: An Encyclopedic Reference Grammar. New Haven CT: Yale University Press. Noonan, M. 1985. Complementation. In Shopen 1985, 42–140. Reh, M. 1985. Die Krongo-Sprache (niino mo-di). Beschreibung, Texte, Wörterverzeichnis [Kölner Beiträge zur Afrikanistik 12]. Berlin: Dietrich Reimer. Rehbein, J., Hohenstein, C. & Pietsch L. (Eds). 2007. Connectivity in Grammar and Discourse [Hamburg Studies in Multilingualism 5]. Amsterdam: John Benjamins. Rennison, J.R. 1996. Koromfe. London: Routledge. Shopen, T. (Ed.). 1985. Complex Constructions [Language Typology and Syntactic Description 2]. Cambridge: CUP. Silverstein, M. 1976. Hierarchy of features and ergativity. In Dixon 1976, 112–71. Watters, D. 1998. The Kham Language of West-Central Nepal (Takale Dialect). Ph.D. Dissertation, University of Oregon, Eugene. Wray, A. (Ed.). 2002. The Transition to Language. Oxford: OUP.

Re(e)volving complexity Adding intonation Marianne Mithun

University of California A fruitful methodology for tracing the development of grammatical complexity has been the examination of centuries of written texts. Yet written documents necessarily remain silent about the prosody of the evolving constructions. An awareness of prosodic patterns can further our understanding of the emergence of complex constructions in several ways. The focus here is on early stages of development of individual constructions within a language, first when prosody is the only indication of complex structure, then when emerging marked constructions are still very young. Processes of development are illustrated with developing complement and relative constructions in Mohawk.

1. Pre-complexity? The past several years have seen avid discussion about whether recursion is an essential feature of language (Hauser, Chomsky, and Fitch 2002; Everett 2005, 2007; Pinker & Jackendoff 2005; Parker 2006; Mithun 2007; Nevins, Pesetsky & Rodrigues 2007; and others). The kind of recursion usually under discussion is hierarchical syntactic structure, in which clauses are embedded inside of other clauses. The central constructions of this type are complementation, where one clause is embedded inside of another as an argument, and relativization, where one clause is embedded as a modifier in an argument of another.

1.1 Complementation Examples of complementation abound in English. Examples (1)a and (1)b both have clausal arguments. The first has a clausal subject: I could deal with it is the subject of be impossible. The second has a clausal object: he cried is the object of he started. (1) English complementation a. It was impossible for me to deal with it. b. And then he started to cry.

 Marianne Mithun

Complement clauses typically have special forms that distinguish them from independent sentences, such as a complementizer like English for or that (It was impossible for me to deal with it), omission of a coreferential subject (He started __ to cry), or a special non-finite verb form (to deal with, to cry). Such structures are not as easy to find in some other languages. In Mohawk, an Iroquoian language indigenous to northeastern North America, all verbs can stand alone as complete, grammatical sentences. (Examples cited here are from spontaneous, speech, generally conversation. The free English translations were provided by the speakers themselves or by others involved in the conversations.)

(2) Mohawk complementation?

a.

Sentential subject: Awenhráthen Joe Deer, speaker p.c. Iáh ki’ teiotòn:’on iah ki’ te-io-at-on-’-on not in.fact neg-neuter.patient-middle-be.possible-inch-stative not in fact was it possible

ken’ niaontié:ren’. ken’ ni-a-ont-k-ieren-’ here partitive-optative-cislocative-1sg.agent-do-prf here I could do (it) ‘It was impossible for me to deal with it.’

b.

Sentential object: Cecelia Peters, speaker p.c. Sok nè:’e tahatáhsawen’ wa’thahséntho’. sok nè:’e ta-ha-at-ahsawen-’ wa’-t-ha-ahsentho-’ so it.is cisloc.fact-m.sg.agt-mid-begin-prf fact-dv-m.sg.agtcry-prf so it is he started (it) he cried ‘And then he started to cry.’

Mohawk is polysynthetic: words, particularly verbs, can contain a potentially large number of meaningful parts. All verbs contain pronominal prefixes referring to their core arguments. Intransitive pronominals refer to one argument, such as w- ‘it’ in ‘it will be impossible’ and -ha- ‘he’ in ‘he cried’. Transitive pronominals refer to two arguments, such as -honwa- ‘they/him’ in wahonwahón:karon’ ‘they invited him’. The transitive pronouns are fused forms: it is often impossible to untangle the agent and patient markers in a transitive prefix. Neuters are not overtly represented unless there is no other argument. The prefix -ha- means both ‘he’ and ‘he/it’: wa-ha-hní:non’ ‘he bought it’, wa’t-ha-hséntho’ ‘he cried’. For this reason, verbs like tahatáhsawen’ in (2)b above could be translated either ‘he started it’ or ‘he started’.

Re(e)volving complexity 

The only obvious relation between the two clauses in each Mohawk sentence above is semantic. In each, a core argument of the first clause is coreferential with the entire second clause. In (2)a, the ‘it’ of ‘It was impossible’ is coreferential with ‘I could deal with it’. In (2)b the ‘it’ of ‘He started to cry’, is coreferential with the event ‘he cried’. Noonan provides a list of semantic types of matrix predicates that appear in complement constructions cross-linguistically.

(3) Semantic types of matrix predicates: Noonan 2007

Utterance predicates say, tell, report, promise, ask … Propositional attitude believe, think, suppose, assume, doubt, deny … Pretence imagine, pretend, make believe, fool, trick into … Commentative/factive regret, be sorry, sad, odd, significant, important … Knowledge and its acquisition know, discover, realize, forget, see, hear … Fear fear, worry, be afraid, be anxious … Desideratives want, wish, desire, hope … Manipulatives force, make, persuade, tell, threaten, let, permit, command, order, request, ask, cause, allow … Modals be able, be obliged, can, ought, should, may … Achievements manage, chance, dare, remember to, happen to, get, try, forget to, fail, avoid … Phasals start, begin, continue, keep on, finish, stop, cease … Immediate perception see, hear, watch, feel … Verbs with all of these meanings appear in Mohawk constructions just like those in (2) above, sequences of fully finite clauses. Additional examples of constructions of these types are in Mithun In press (a) and In press (b). The fact that the counterparts of English complement clauses show no special dependent forms is not altogether surprising, given the overall structure of the language. Since all verbs contain obligatory pronominal prefixes and are finite, dependent clauses could not be distinguished by ellipsis of a coreferential argument or dependent inflectional forms.

 Marianne Mithun

1.2 Relativization Examples of relative clauses are also not difficult to find in English. These are clausal modifiers of an argument of a higher matrix clause. In (4), the children is modified by the clause (they) came here.

(4) English relativization Maybe the bus brought the children [that came here].

This English relative clause differs from main clauses in two ways. It is introduced by the relative pronoun that, and it is missing a regular pronominal or lexical subject. The English sentence in (4) was actually the free translation of the Mohawk sentence in (5), part of a conversation. The Mohawk shows neither a relative pronoun nor omission of the coreferential argument. (5) Mohawk relativization? Charlotte Bush and Josephine Horne, speakers p.c. CB Tóka’ ki’ nè:’ ne ki: iakoia’takarénie’s toka’ ki’ nè:’ne ki: iako-ia’t-a-kareni-e’s maybe just it is this indefinite.pat-body-lk-transport-distr maybe just it is this it bodily transports one here and there ‘Maybe the bus

thotiia’ténha’ t-hoti-ia’t-enha’ cisloc-m.pl.pat-body-carry it bodily carried them here brought them, didn’t it,

wáhi’, wahi’ tag didn’t it

JH Mm. CB

ki: ratiksa’okòn:’a, ki: rati-ksa’=okon’a this m.pl-be.a.child=distr these children the children

JH Mm CB

thoné:non t-hon-e-n-on cisloc-m.pl.pat-go-dir-stative they have come that came here.’

kèn:’en. kèn:’en here here

Re(e)volving complexity 

The Mohawk construction does show some semantic characteristics typical of relative clauses. The two clauses ‘Maybe the bus brought the children’ and ‘they came here’ share an argument, the children. It is often maintained that subordinate clauses, including relative clauses, represent presuppositions rather than assertions. The last clause ‘they came here’ conveys a presupposition. The two speakers had been standing together on the front porch the day before, watching the children arrive.

1.3 The prosodic dimension Sequences like those in (2)a, (2)b, and (5) above are pervasive in Mohawk. They could be taken as evidence that the language lacks syntactic complexity. They appear to consist of strings of independent sentences with no special relationship except a semantic one, perhaps one that is only inferred. If, however, we move beyond the printed word to a consideration of sound, additional structure emerges. Prosody is generally understood as some combination of pitch, intensity, and timing. In the investigation of complex structures, the most significant of these features is pitch movement or intonation. A typical example of simple sentence intonation can be seen in (6), a sequence of two separate sentences. (6)

Mohawk sequence of sentences: Watshenní:ne Sawyer, speaker p.c. Khe tóka’ ioiánere’. khe tóka’ io-ianer-e’ I.think maybe neuter.patient-be.good-stative I think maybe it is good ‘I think maybe it would be good.

Tóka’ né: iakenenhrénhawe’. toka’ ne: i-a:-ke-nenhr-enhaw-e’ maybe it.is translocative-optative-1sg.agent-group-carry-prf maybe it is I should group take there I should take the whole group in (to the doctor).’

The intonation can be seen in the pitch trace below. The first sentence began on a high pitch then fell more or less continuously to the end. The two pitch bumps occurred on the two stressed syllables. The second sentence began with a pitch reset, up to the same high pitch as the beginning of the first sentence. It ended with a similar final fall. Here too there were two bumps of high pitch, on the two stressed syllables. These two sentences were separated by a brief pause, a frequent but by no means necessary characteristic of the boundaries between prosodic phrases. The overall pitch contours are criterial.

Pitch (Hz)

 Marianne Mithun

200 150 ‘Khe tóka’ioiánere’. I think it might be good.

100 0

Tóka’ ne: iakenenhrénhawe’. Maybe I should take the group in. 2.35683

Time (s)

Grammatically, this sentence is very similar to ‘It was impossible for me to deal with it’ in (2)a. The first clause in each contains just one argument (‘It would be good’, It was impossible’. The pronoun ‘it’ refers to the fact stated in the second clause: ‘I should take them in’, ‘I could deal with it’. The two show very different intonational patterns, however, as can be seen by comparing the pitch trace above to that below. (2)a. ‘It was impossible for me to deal with it.’

Pitch (Hz)

150

100 70 50

0

Iáh ki’ teiotò:’on

ken’ niaontié:ren’.

It is just not possible

I could just do it there. Time (s)

1.51365

The intonation in (2)a reflects integration of the two clauses into a single prosodic structure. There is a steady fall in pitch from the beginning of the first clause to the end of the second, with bumps on the stressed syllables at successively lower pitches. The first clause did not show a full terminal fall in pitch, and the second did not begin with a pitch reset. A similar contrast can be seen by comparing sequences of sentences with constructions translated with English sentential objects. Example (7) consisted of two independent sentences, each constituting a separate prosodic phrase. (7)

Separate sentences: Sonny Edwards, speaker p.c. Sok iaken’ tahontáhsawen. so hearsay it started ‘So then, it seems, it started.

Wa’tkanón:wáhkwe’ ki: awèn:ke. it started swirling this water The water started swirling around.’

Re(e)volving complexity 

The first sentence ‘So then it started.’ ended with a terminal fall. In the second sentence, the pitch of the first stress syllable -nón- was as high as the first stressed syllable of the preceding sentence.

Pitch (Hz)

150 100 70 50 Sok iá:ken’ tahontáhsawen’. Then they say it started. 0

Wa’tkanón:wahkwe’ ki: awèn:ke. It swirled this water. 3.34948

Time (s)

This prosodic structure contrasts with that of (2)b. (2)b. ‘So then he started to cry.

Pitch (Hz)

300 200 150 100 0

Sók nè:’etahatáhsawen’

wa’thahséntho’.

So the he began it

he cried.

Time (s)

2.2146

Here the two clauses were integrated under one overall intonation contour, with no full terminal fall until the end of the last word. (The final syllable -tho’ of ‘he cried’ does not come through well on the pitch trace, due to devoicing, but it is audible.) There was a regular decrease in pitch (declination) from one stressed syllable to the next, that is, from the stressed syllable of Sók ‘then’, to the stressed syllable -táh- of ‘he started it’ then finally to the stressed syllable -sént- of ‘he cried’. Constructions translated with English relative clauses are characterized by similar prosodic integration. The passage in (8) was later translated by the speaker as a sequence of two independent sentences in English: ‘He was a beautiful little dog. His name was Butch.’ The first sentence ends with a terminal fall in pitch, the second begins with a pitch reset, and the two are separated by a substantial pause. (8)

Separate sentences: Watshenní:ne Sawyer, speaker p.c. Ranahskwiióhne’ ken’k nì:ra è:rhar. ra-nahskw-iio-hne’ ken’=k ni-hr-a ehrhar. m.sg.agt-animal-be.good.-past little=only prt-m.sg.agt-be.sized dog he was a good animal just small so he was sized dog ‘He was a beautiful little dog.

 Marianne Mithun

Butch ronwá:iatskwe’ Butch ronwa-iat-skwe’ Butch 3pl/m.sg-call-past.hab Butch they called him His name was Butch.’

Pitch (Hz)

300 200 150 100 Ranahskwiióhne

70

50 He was a beautiful animal

ken’k nì:ra è:rhar.

Butch ronwá:iatskwe’.

he was a little dog.

Butch onecalled him.

0

3.88934

Time (s)

It is significant that this speaker did not choose to render it with a relative clause ‘He was a beautiful little dog named Butch.’ The passage contains two asserted ideas, each presenting new information of its own. The grammatical structure of (8) is similar to that of sentence (5) seen earlier: ‘Maybe the bus brought the children that came here’. In both, the second clause provides additional information about the noun immediately before it. The two contrast in their prosodic structure, however. While (8) was pronounced as two distinct prosodic phrases, each with its own terminal fall in pitch and a pause between, (5) was pronounced under a single overall prosodic contour. The pitch moved from a high at the beginning of the first clause to a full terminal fall at the end of the second clause. (5) ‘Maybe the bus brought the children that came here.’ 300

Pitch (Hz)

200 150 100 70 50 30 20

Tóka’ki’ nè:‘ ne ki: iakoia’takarénie’s thotiia’ténha’ wáhi, Maybe the bus 0

Mmm

brought them here

ki: ratiksa’okòn:’a, these children

Time (s)

Mmm

thoné:non kèn:’en. they havecome here. 6.98921

The larger contour shows internal structure. Constituent sub-phrases are set off by vertical lines on the pitch trace. The second sub-phrase began with a partial pitch reset, not as high as the beginning of the sentence as a whole. Successive sub-phrases were separated by brief pauses and slight pitch falls. These units were evidently perceived by the other speaker, who inserted back channel responses (Mmm) at just these points. This prosodic pattern, one overall prosodic contour made

Re(e)volving complexity 

up of small constituent sub-contours, could be characterized as recursive structure, the embedding of one type of structure inside another of the same type.

1.4 Syntactic and prosodic structures The prosodic integration of constructions like those in (2)a, (2)b, and (5) reflects a kind of cognitive organization similar to that reflected by syntactic integration. The fact that we find prosodic structure without substantive syntactic structure suggests that prosodic structuring might, at least in some cases, precede syntactic structuring. But as Bolinger (1984, 1989) pointed out early on, prosodic and syntactic structure are not necessarily isomorphic. I start with a claim and a disavowal. The claim is that intonation is autonomous and one can speak of intonational subordination without reference to the segmental side of language. The disavowal is that intonation has any direct connection with subordination in syntax, however this is to be defined. Syntax nevertheless benefits handsomely from the games that intonation plays with it. I see anything that is tributary to something else as subordinate to it. In syntax this means not only the classical dependent clauses in relation to main clauses, but also their reduced counterparts … In Gestalt terms, what is superordinate is the figure; what is subordinate is all or part of the ground. (Bolinger 1984:401)

Prosodic and syntactic structure often go hand in hand, but they can diverge to convey different aspects of the message. In the complex Mohawk constructions seen in (2) and (5), what is interpreted as the matrix clause always occurs first, followed by what is interpreted as the subordinate clause. This is a regular pattern. Each shows an overall fall in pitch as well: each stressed syllable is lower than the preceding one. The matrix clause shows higher pitch than the complement or relative clause. The highest pitch is not always on the matrix clause, however. Consider the subject complement construction in (9). (9) Subject complement: Joe Awenhráthen Deer, speaker ‘(If I’m still in good health,) it should be possible for me to make my garden a little bigger.’ ó:nen ki’ enwá:tón’, then just it will be possible then I might be able kwah ostòn:ha enkathehtó:wanahte’ quite a little will I field enlarge for myself to make my garden a little bigger.’

nòn:wa. this time

 Marianne Mithun 150 100 70 50

ó:nen ki’ enwá:ton’ Now it will be possible

kwáh ostòn:ha enkathehtó:wanahte’ nòn:wa quite a bit I will garden enlarge this time

0

3.00698

What would be identified as the syntactic matrix clause ‘then it will be possible’ was spoken with significantly lower pitch than the following clause ‘I’ll enlarge my garden’, which would be identified as the syntactic complement. (Particles, like kwah ostòn:’a ‘quite a bit’, do not generally receive strong stress in context.) This is not an isolated example. The sentential object construction in (10) shows a similar pattern. The speaker was describing what she had just seen in a film. (10) Object complement: Kaia’titáhkhe’ Jacobs, speaker p.c. [‘I believe it was early in the morning, because] wakathón:té’ . . . I heard (it) ‘I heard

um . . .

kítkit rá:tsin wa’thohén:reh’te’. chicken male he yelled a rooster crow.’

200 150

wakathón:té’ I heard 0

um

kítkit rá:tsin wa'thohén:rehte’. rooster he yelled 4.44662

Despite the pause following the first clause, the prosodic integration of this construction is still clear. The initial clause ‘I heard it’ did not end with a full terminal fall. It shows a special phrase-medial tone structure. Stress is basically penultimate in Mohawk, except that certain epenthetic vowels do not enter into the determination of stress. (These vowels were added after the basic penultimate stress pattern ofProto-Northern-Iroquoian was established.) Stressed syllables generally are distinguished by high or rising pitch (á, á:) or rising then plunging pitch (à:). Historically

Re(e)volving complexity 

open syllables are lengthened. If stress falls on an open penultimate syllable, and the word is not phrase-final, the pitch rise continues into the ultimate syllable. Thus the word ‘I heard it’ is wakathón:te’ at the end of a phrase. (The digraph on represents a nasalized vowel.) If another word follows it in the same prosodic phrase, it is wakatón:té’. The verb in (10) showed this non-final form, indicating that the speaker intended to continue immediately with the next clause within the same prosodic phrase. What is significant about this example is that the complement clause ‘a rooster crowed’ was significantly higher in pitch than the matrix ‘I heard it’. In both (9) and (10) the main information is carried by the syntactically embedded clause. The matrix verb ‘it will be possible’ in (9) is serving a modal function, ‘I might enlarge my garden’. The complement clause did not convey presupposed information: the news here was about enlarging the garden, not about possibilities. The matrix verb ‘I heard it’ in (10) is serving an evidential function. Again the complement clause was not presupposed: the news was not the act of hearing but the rooster crowing. Other authors have noted the mismatch between the syntax of complement constructions and information in other languages. Describing English, Thompson writes: The standard view of complements as subordinate clauses in a grammatical relation with a complement-taking predicate is not supported by the data … Rather, what has been described under the heading of complementation can be under-stood in terms of epistemic/evidential/evaluative formulaic fragments expressing speaker stance toward the content of a clause. (Thompson 2002:125)

Verhagen comes to a similar conclusion about written Dutch. Complementation constructions have the primary function of instructing the addressee of an utterance to coordinate cognitively — in a way specified by the matrix clause — with another subject of conceptualization in construing the object of conceptualization (the latter being represented by the complement clause) and not that of representing an object of conceptualization. (Verhagen 2005)

The prosodic structure of examples like those in (9) and (10) can affect the further development of grammatical structures. Verbs like ‘it is possible’ and ‘I heard’ are just the kinds of words that tend to be reduced over time into auxiliary verbs, evidential particles, clitics, and affixes. A number of such developments can be seen within Mohawk itself. Mohawk contains, for example, a regular verb iá:ken’, based on the root -en- ‘say’. (11)

Verb -en- ‘say’ iá:ken’ iak-en-’ indefinite.agent-say-stative ‘one says, they say, people say’

 Marianne Mithun

This verb can still be used as the matrix clause in complement constructions, but it occurs much more often as a hearsay evidential. In this use it is typically reduced in form, losing the stress and length. As a matrix verb it is pronounced iá:ken’, but as an evidential it is iaken’. It also shows some freedom of movement. Its status as an emerging particle can be seen in (12) ‘They just took him up there’. Here it was pronounced with little stress or length and was embedded inside of the clause, occurring after both ‘there’ and ‘just’. The point of the sentence is not that ‘one just said it there’, but rather that ‘they just took him there’. (12)

Hearsay evidential: Josephine Horne, speaker p.c. Thó ki’ iaken’ iahonwaia’ténhawe’, tho ki’ iak-en-’ i-a-honwa-ia’t-enhaw-e’ there just indefinite-say-stative trloc-fact-3.m.pl/3.m.sg-bodycarry-prf there just hearsay they bodily took him ‘They apparently just took him up there … ’

Pitch (Hz)

200 150 100 70 0

Thó ki’ iaken’

iahonwaia’ténhawe’,

There they say

they took him Time (s)

1.70667

2. Young marked complement constructions Givón (2002, 2005, in this volume), and Heine and Kuteva (2007) identify two principal paths by which subordination develops. There are cross-linguistically two main ways in which clause subordination arises: either via the integration of two independent sentences within one sentence or via expansion, that is, the reinterpretation of a thing-like (nominal) participant as a propositional (clausal) participant. (Heine 2008ms:1)

Heine and Kuteva’s ‘integration’ corresponds to Givón’s ‘clause chaining’, and their ‘expansion’ to his ‘embedded verb phrase complementation’ or ‘nominalized V-COMP’. Diessel, who originally coined the terms ‘integration’ and ‘expansion’, found that in children’s acquisition of their first languages, ‘complement and relative clauses evolve

Re(e)volving complexity 

via clause expansion’ (Diessel 2005:4, cited in Heine 2008ms:1). Complex sentences appeared later than simple sentences in the speech of children. A consideration of prosody allows us to examine the effects of these two processes and their interaction more closely. The examples of Mohawk complementation seen so far appear to reflect simple integration. Some sequences of grammatically independent sentences, in which an argument of the first clause is coreferential with the entire second clause, are now integrated prosodically. There are, however, additional complement constructions that show more than simple integration.

2.1 Complementation with demonstratives Mohawk contains two demonstratives, a proximal kí:ken ‘this, this one, these’ and a distal thí:ken ‘that, that one, those’. They are often shortened: to kí: and thí: respectively. Like demonstratives in other languages, they serve to locate a referent in space or time, in the linguistic or extralinguistic context. (13) Proximal demonstrative: Margaret Lazore, speaker Kthontaiawénhstsi, all of a sudden ‘On the spur of the moment,

wahonterihwahserón:ni’ ahatiiá:ken’ne’ they made an agreement they should go out they decided to go out this Saturday.’

(14) Distal demonstrative: Margaret Edwards, speaker Thí:ken orokwáhsa’ entehsié:na. that chain you will grab it ‘You’ll grab that chain.’ (15) Demonstratives: Lazarus Jacob, speaker É: ì:reht thí:ken; away may he move that ‘Get that guy out of the way;

enhahétkenhte’ kí:ken ne case. he will make it bad this the case. he’ll ruin this case.

Kí:ken sò:tsi this too much This guy

rahnekakà:stha’. he habitually liquid overdoes drinks too much.’

ki: entákta’. this Saturday

 Marianne Mithun

The demonstratives may occur on their own, as in (15) ‘that (guy)’ and ‘this (guy)’, or in combination with a coreferential nominal, as in (13) ‘this Saturday’, (14) ‘that chain’, and (15) ‘this case’. They can appear with possessed nouns and proper names. Interestingly, they can also precede clauses. (16) Complement with kí:ken ‘this’: Lazarus Jacobs, speaker Rérha’ enhoió’ten’ kí:ken enhshakoia’totáhsi’ ratitshihénhstatsi he intends he will work this he will expose them priests ‘He intended [to work [to expose the priests]].’ (17) Complement with thí:ken ‘that’: Joe Tiorhakwén:te’ Dove, speaker Tóka’ ken enhsehià:rake’ thi: maybe Q you will remember it that wahshakonahskwawíhon wahi’. he gave away livestock to various people you know

‘Maybe you remember [that he gave away livestock], right?’

The presence of demonstratives before complement clauses indicates that these clauses are conceived of as referring expressions. The construction could be viewed as the result of expansion: determiner phrases originally filled with a demonstrative plus noun were expanded to allow a demonstrative plus clause. This would parallel the path of development traced by Diessel in children’s acquisition of English complement constructions. A closer look at a common discourse structure in Mohawk suggests an alternative scenario. The sentence in (16) was packaged prosodically as one large unit with internal structure consisting of three sub-phrases, each a clause: ‘he intends it’, ‘he will work for this’, and ‘he will expose the priests’. The second and third sub-phrases each began with a partial pitch reset on the first stressed syllable, but these pitches were not as high as the initial pitch on the matrix verb ‘he intends’. There was no full fall until the end of the third and final clause.

Pitch (Hz)

(16) ‘He intended [to work [to expose the priests]].’: Lazarus Jacob, speaker 300 200 150 100 Rérha’ 70 50 He intends 0

enhoió’ten’ ki:kén: he will work this

enhshakoia’totáhsi’ ratitsihénhstatsi. he will expose the priests. Time (s)

4.44082

Re(e)volving complexity 

The entire construction was integrated under one overall prosodic contour, but a break can be heard between the second and third clauses. Interestingly, the break followed the demonstrative kí:ken ‘this’. In this example it took the form of lengthening on the final syllable of the demonstrative. This is not a terminal contour: the rise in pitch on the final syllable of kí:kén: indicates that more is to follow. It is the phrase-medial intonation pattern described earlier in section 1.4. The break between matrix and complement clauses is even more pronounced in (17). The two clauses were separated by a substantial pause, visible in both the pitch trace and the waveform. Here again, the demonstrative was grouped prosodically with the matrix rather than the complement. (17) ‘Maybe you remember that [he gave away livestock], right?’ 0.6526 0 −0.6293

Pitch (Hz)

150 100 70 50 30

0

Tóka’ ken enhsehià:rake’ thi:

wahshakonahskwawíhon wahe’.

Maybe you remember that

he gave away livestock you know. Time (s)

3.43075

(These breaks are not pauses for word searches; such structures show different prosody.) As been pointed out in Pawley and Syder 1975; Pawley 2000, and Chafe 1979, 1982, 1987, 1994, spontaneous speech is typically not produced in a continuous stream. Speakers regulate the flow of information such that, in essence, they introduce just one new idea at a time per intonation unit or prosodic phrase. The new idea might be the introduction of a new participant, action, time, place, or other new or significant item of information. Chafe describes this structure as follows. The fact that in the end we are left with few if any cases in which there are two or more separately activated new ideas within the same intonation unit suggests the hypothesis that an intonation unit can express no more than one new idea. In other words thought, or at least language, proceeds in terms of one such activation at a time, and each activation applies to a single referent, event, or state, but not to more than one. (Chafe 1994:109)

 Marianne Mithun

Pawley similarly observes that there is a fundamental limit on cognitive processing, which concerns the number of units of new information that can be manipulated in a single focus of consciousness . . . Two factors place time constraints on speakers’ strategies for formulating speech in meetings (face-to-face encounters): first, the social context, which usually places a premium on packaging talk for a fast ride; and second, biological limits on what the mind can do at speed’ (Pawley 2000:164, 165).

There is of course variation in the duration of prosodic breaks between intonation units, both across speakers and within the speech of single speakers. The management of information flow can be seen in the passage in (18). The passage could be translated ‘The late Kahonwinéhtha’ always used to go visit her daughter Konwahsé:ti in New York City in the wintertime.’ The Mohawk is arranged by intonation unit: each line represents a prosodic phrase. (18) One new idea at a time: Joe Awenhráthen Deer, speaker Ne: ki’ thi:ken … it is anyway that

akokstenhkénha late old lady

Kahonwinéhtha’ (name)

thó ienienatahré:nawe’ there she used to visit way over there

(Én:. yes)

tiótkon’s thi n-akohserà:ke always that the wintertime

enienatà:ra’. she’ll visit.

Konwaièn:’a thí:ken her daughter that

Konwahsé:ti (name)

tho ses nonkwa(ti) tienákere’ there formerly over there there she resides

Kanón:no. New York City.

‘The late Kahonwinéhtha’ always used to go visit her daughter Konwahsé:ti in New York City in the wintertime.’

Re(e)volving complexity 

Each phrase introduced a new idea. The first shifted the topic of conversation. The second identified the new main character by name, old Kahonwinéhtha’. The third stated her basic activity ‘she used to visit’. (The fourth was the response of another speaker.) The fifth specified the time of the visits. The sixth introduced another character, the daughter. The seventh identified the daughter by name. The eighth brought up her residence. The ninth identified the location by name. This example also illustrates a common Mohawk rhetorical pattern. Demonstratives are often used as place holders, indicating in one phrase that further details are to follow. In the first line of (18) the demonstrative thí:ken ‘that’ establishes a referent that is further identified in the following phrase as Kahonwinéhtha’. In the sixth line the same demonstrative promises further information about the daughter. The construction in (17), ‘Maybe you remember that … he gave away livestock’, was of the same type. Sequences of separate sentences following this pattern are common in Mohawk. The sentences are grammatically and prosodically independent. Neither presents presupposed information. (19) Two sentence sequence with demonstrative: Watshenní:ne’ Sawyer, speaker p.c. Eniakwaterohrókha’ kí:ken; … we will go to watch this ‘We would go watch this;

tewa’á:raton tahonhthénno’ke’. it is net attached they (males) would play ball the men would play lacrosse.’ 0.5591 0 −0.7582

3.91837

0

Pitch (Hz)

150

100

0

Eniakwaterohrókha’ kí:ken.

Tewa’á:raton tahonhthénno’ke’.

We will watch this.

They would play lacrosse. Time (s)

3.91837

As can be seen from the waveform and the pitch trace, the two clauses were separated by a long pause. The first clause ended with a partial fall, but the second showed a

 Marianne Mithun

complete pitch reset. Such structures are likely sources for complement constructions such as ‘He intended to work to expose the priests’. The existence and frequency of examples like (19) suggest a slightly different path of development for the Mohawk complement constructions with demonstratives than that observed in children acquiring their English counterparts. The Mohawk constructions may have originated in a discourse pattern of elaboration consisting of a sequence of clauses, in which a demonstrative in one sentence announced that fuller identification of a referent was to follow in the next: ‘We would watch this. They would play lacrosse.’ The cognitive integration of the two ideas came to be mirrored by prosodic integration. The result was a complement construction containing a demonstrative, where the demonstrative is grouped prosodically with the first clause rather than the second.

2.2 Complementation with articles Mohawk contains another marked complement construction. Lexical nominals are not obligatorily marked for definiteness, but there is a particle ne which can indicate that the speaker judges the referent to be identifiable to the listener from previous mention. It is much like the English definite article. An idea of its function can be gained by comparing two sentences from an anecdote. The speaker was describing the time a man had been caught inadvertently doing something illegal. One of his relatives urged the action in (20). There was no ne before the word ‘lawyer’. (20) No ne: Watshenní:ne’ Sawyer, speaker p.c. Ó:nenk tsi tehari’wakénhahs right now he argues matters entshitewaia’tatshén:ri’. we will look for him ‘We have to find a lawyer right away.’ The family did locate a lawyer, and the case went to trial. The next mention of the lawyer was the sentence in (21). This time the word ‘lawyer’ was preceded by ne. The lawyer was now identifiable from the preceding discussion. The word for ‘judge’ was also preceded by ne, though this was the first mention of him. He was assumed to be identifiable from the courtroom scenario. (21)

Ne: Watshenní:ne’ Sawyer, speaker p.c. Ah khare’ ó:nen ki: ia’káhewe’ ne tekari’wakénhahs ah so then this it arrived there the he argues matters ‘So then this time the lawyer

tanon’ ne shakorihwénhtha’ and the he decides people’s matters and the judge

Re(e)volving complexity 

wa’thonwaia’tò:rehte’ they judged him brought him to trial, ki: X. this name this Mr. X.’ The particle ne can co-occur with demonstratives, as in kí:ken ne case ‘this case’ in (15) above and with proper names. It is not, however, obligatory, even when the referent is identifiable. The sentence in (22) occurred sometime after the sentence seen earlier in (15): ‘You’ll grab that chain.’ (22)

No particles: Margaret Edwards, speaker, p.c. Tahaié:na’ orokwáhsa’. he grabbed it chain ‘He grabbed the chain.’

Interestingly, ne can appear before clauses in complement constructions. (23) Complement with ne ‘the’: Cecelia Peters, speaker p.c. Iakwate’niénhtha’ ne akwé:kon onkwehón:we we habitually try it the all real person a:iakwatewennón:tahkwe’. we would our word stand with it ‘We try [to speak only Indian].’ The sentence in (23) was uttered with the same prosodic integration seen in other complement constructions. The complement ‘to speak only Indian’ was embedded prosodically inside of the sentence ‘We try to speak only Indian’. There was no full fall in pitch until the end of the whole. The first clause ‘we try’ ended in only a partial fall, and the second clause, translated as the complement, began with only a partial pitch reset.

Pitch (Hz)

300 200 150 100

Iakwate’niénhtha’ We try 0

ne akwé:kon onkwehón:wé a:iakwatewennón:tahkwe’. we would say everything in Indian. Time (s)

3.54685

The prosody of complement constructions with ne ‘the’ like that in (23) differs in an important way from that of demonstrative constructions. While the

 Marianne Mithun

demonstratives are grouped prosodically with the preceding matrix clause, the particle ne is grouped with the following complement clause. This particle cannot introduce independent sentences: the sequence ne akwé:kon onkwehón:we a:iakwatewennón:tahkwe’ ‘the we all speak Indian’ makes no sense. The ne complement constructions could thus not have developed according to the scenario proposed for the demonstrative complement construction: simple prosodic integration of two sentences. Several other scenarios are possible. The ne construction could have come about through elaboration of the type described by Diessel for children learning English. Speakers could have generalized Det N phrases identifying arguments (‘the chain’) to Det S phrases identifying clausal arguments (‘the our speaking only Indian’). Alternatively, it could have originated as an unmarked complement construction like those seen in section 1: ‘We try it; we say everything in Indian’. Speakers then could have extended the definiteness marking used with nouns to clauses also serving as referring expressions. It might be tempting to assume that the Mohawk demonstratives and definite article have now attained the status of complementizers, much like English that. There is evidence that they are not yet at that point. All three still mark the same semantic distinctions with clauses that they mark with lexical nominals. The demonstratives distinguish proximal from distal situations: events or states that are near or remote in space, time, or discourse. In ‘Maybe you remember thí:ken (‘that’) he gave away livestock’, the speaker was talking about a remote time, during the Depression of the 1930’s. In ‘He intended kí:ken (‘this’) to work to expose the priests’, the speaker was referring to the central topic of the current conversation, a lawsuit over land ownership. The particle ne ‘the’ still marks exactly the same distinction before clauses that it marks before lexical nominals: identifiability of events and states. It is not used before clauses that introduce brand new information. Furthermore, a demonstrative and the article ne can occur together before a clause. As expected, the demonstrative is grouped prosodically with the matrix clause before it, while the article is grouped with the complement clause after it. (24)

Coccurrence of demonstrative and article: Cecelia Peters, speaker p.c. Kè:iahre’ thi: I remember that ‘I remember that

ne s ne: wakon’éskwani’ the past it is I like it I used to like it tsi náhe’ eh niiohtòn:ne’. long ago there so it was remotely the way it was long ago.’

Re(e)volving complexity 

Pitch (Hz)

200 150 Ke:iahre’ thi: I remember that

100

nes ne: wakon’eskwani’ the I used to like it

0

tsi nahe’ eh niiohton:ne’. it used to belike that 3.20435

Time (s)

3. Young marked relativization Examples were seen in section 1 of relative-like constructions marked only by prosodic integration. Mohawk also contains some young marked relative constructions.

3.1 Relativization with demonstratives The sentence in (25) appears to contain a standard relative clause. (25)

Demonstrative: Joe Tiorhakwén:te’ Deer, speaker Nahò:ten’ na’ thí:ken wà:kehre’ enkehià:rake’? what now that I wanted I will remember ‘What was it now that I meant to remember?’

The clause ‘I meant to remember something’ is a presupposition. As can be seen from the pitch trace, the full sentence was uttered under a single overall intonational contour. There was no full fall in pitch until the very end. (The slight rise in pitch on the stressed syllable of the final verb ‘remember’ is due to the tone, written with a grave accent, which consists of an initial higher rise followed by a very steep fall.) (25) What was it now that I meant to remember? 150

Pitch (Hz)

100 70 50 30 20 15

0

Nahò:ten’ na’ thí:ken,

wà:kehre’ enkehià:rake’?

What was that

I meant to remember? Time (s)

2.52517

The diachronic pathway generally assumed to underlie relative clauses in languages like English is the following (Heine & Kuteva 2007).

 Marianne Mithun

(26) There is the car; that (one) I like > There is the car [that I like]. A closer look at the prosody of (25) suggests that this is not the pathway by which the Mohawk construction developed. This sentence was uttered with a pause between the two clauses. The demonstrative thí:ken ‘that’ was grouped with the first clause. There is in fact often a significant prosodic break between the demonstrative and the following clause in such constructions. An example with the other demonstrative kí:ken ‘this’ is in (27). Again the transcription is arranged so that each line represents a separate prosodic phrase. (27) Larger demonstrative construction. Joe Awenhráthen Deer, speaker Nòn:wa kí:ken, now this

òn:wa’k wahonwaia’táta’ thetèn:re’, just now they buried him yesterday

Eddie,

Eddie Delaronde,

ne s ne: it is formerly the

rake’níha akwas my father really

akì:ron tsi ki’ ní: ne: tehiatatshnié:nenhskwe’ wáhi. I’d say that in fact myself it is they two used to help each other tag

‘This guy [they just buried yesterday],

Eddie, Eddie Delaronde, he and my father used to just help each other out, you know.’

Pitch (Hz)

150

100 70 50

Nòn:wa kí:ken, òn:wa’k wahonwaia’táta thetèn:re’, Eddie, Eddie Delaronde, Now this, 0

they just buried him yesterday Time (s)

ne s ne:

rake’níha akwas uh

akì:tehiatatshnié:nenhskwe’

used to be

my father

they used to help each other 8.80036

The relative construction, a topicalized phrase, is contained in the first two intonation units: ‘Now this guy they just buried yesterday’. The pitch trace for just this portion is repeated below, along with the waveform. There is a consistent decrease in pitch over successive stressed syllables in the phrase. (The plunge on the initial syllable is from the falling tone.)

Re(e)volving complexity 

(27) ‘This guy they just buried yesterday … ’ 0.9059 0

Pitch (Hz)

−0.5522 200 150 100 70 50 30 20

Nòn:wa kí:kén:, Just now this, 0

òn:wa’k wahonwaia’táta’ thetèn:re’, they just buriedhim yesterday Time (s)

2.85025

These structures do not pattern like that proposed to underlie English relative clauses: There is the car; that (one) I like. The demonstratives are grouped prosodically with the matrix rather than the relative clause. Furthermore, demonstratives are the only words that can occupy this position: ordinary lexical nouns never appear here. The rhetorical structure seen in the previous section, illustrated in example (18) ‘The late Kahonwinéhtha’ always used to go visit her daughter Konwahsé:ti in New York City in the wintertime’, again suggests a different pathway of development. It is likely that the complex constructions (26) and (27) sprang from a similar source, in which a demonstrative in one intonation unit serves as a place holder promising further elaboration in the next. Heine and Kuteva note that as demonstratives develop into relative pronouns, ‘desemanticization leads to a loss of the spatial deixis of the demonstrative’. (2007:225) The Mohawk demonstratives in constructions like (26) and (27) retain their deixis, distinguishing distance in space, time, or discourse. In ‘this guy they just buried yesterday’, the proximal kí:ken ‘this’ emphasized the proximity in time, ‘just yesterday’. In ‘What was it that I meant to remember’, the distal thí:ken ‘that’ referred to a moment the speaker could no longer remember well. The difference is of course relative, not absolute. The burial had taken place the day before, while the thought of something to remember could have occurred to the speaker earlier the same day.

3.2 Relative constructions with né:ne and né: Mohawk contains another relative-like construction marked with the particle né:ne or né:. Like the relative construction with demonstratives, this one does not usually include a lexical noun, though on occasion there may be one.

 Marianne Mithun

Unlike the demonstratives, these particles are grouped prosodically with the relative clause. (28) Né:ne without noun: Lazarus Jacob, speaker Né:ne roió’tehkwe’, the one he was working ‘The one that was working,

kí:ken X, this X

né: róhson ki: [kahiatónhsera’] that.one he has painted this written thing he’s the one who had written this paper.’

Pitch (Hz)

(Other speaker) [A:] ‘Ah.’ 200 150 100 70

Ne:ne roió’tehkwe’, The one working

50 0

kí:ken X this X Time (s)

né: róhson ki: kahiatónhsera’/ A: he made that paper. / Ah. 4.11574

A construction with lexical noun is in (29). (29) Né: construction after noun: Kahentoréhtha Marie Cross, speaker p.c. Wà:kehre’ nahò:ten’ ne: kèn:’en ratstà:ne’ kí:ken, I thought what that here he is going to use this ‘I thought, “What is he going to do with this thing,

raksà:’a boy the boy

né: ki: thó: ranonhwétsha’. that this there he sleeps that sleeps there?”’.

The pitch trace shows that the preceding noun did not end in a final fall, as it would at the end of a prosodic sentence. The né: clause began with a pitch reset, however, much as it would in a separate sentence.

Re(e)volving complexity 

Pitch (Hz)

700 500 300 200 150 100 70 50

Wà:kehre nahò:ten ne kèn:’en ratstà:ne kí:ken

raksà:'a

né: ki: thó ranonhwétsha’.

I thought What's he going to do with this thing,

boy

That’s the one that sleeps there.

0

Time (s)

4.41179

Constructions with né:ne or né: function much like restrictive relatives. Their likely source is still pervasive in modern Mohawk, in independent sentences like those in lines two and three below. (30) Independent ne: sentences: Josephine Horne and Charlotte Bush, speakers p.c.

JH Ónhka’ iáh teiakohthotá:ton? who not has one heard ‘Who didn’t agree?’

CB Né:ne ro’níha kí:ken. that is his father this ‘It was his father.’

JH A:, né: wahánhe’. ah that one he forbade it ‘Oh, he’s the one that forbade it.’

The mechanism behind relative constructions like those in (28) and (29) is apparently simple integration, still in progress.

4. Conclusion Consideration of prosodic structure can greatly enrich our understanding of the pathways by which complex constructions can develop. It has been observed that some languages appear to lack complexity altogether; all information is simply expressed in strings of syntactically independent sentences. Mohawk might appear to be such a language. An examination of the prosody of such strings, however, reveals structural complexity in another dimension. The existence of complex prosodic structures without morphosyntactic markers indicates that at least in some cases, the emergence of prosodic structuring might precede that of syntactic structuring. It is of course important to note that prosodic structure is not necessarily a straightforward precursor to syntactic structure. Each kind of structure can indicate distinctions the other

 Marianne Mithun

does not. It has sometimes been assumed, for example, that matrix clauses are always asserted, while subordinate clauses are presupposed. Examination of spontaneous speech indicates that syntactically subordinate clauses are in fact often not presupposed, and prosody can mark the difference. In the absence of an ancient, detailed written record, we cannot know for certain exactly how the modern Mohawk complement and relative constructions developed. Some insight can be gained from the comparison of related constructions at different stages of development coexisting within the modern language. The examination of modern speech offers an advantage not available from written records: the possibility of comparing prosodic structures. The prosodic patterns of the Mohawk constructions suggest certain pathways of development that differ somewhat from those proposed for European languages. In addition to those marked by prosody alone, Mohawk contains both complement and relative constructions marked by demonstratives. Unlike their English counterparts, the Mohawk demonstratives are grouped prosodically with the preceding matrix clause rather than the following dependent clause. This pattern suggests an origin in a pervasive discourse pattern used to manipulate the flow of information through speech. Each prosodic phrase or intonation unit introduces just one significant new idea. Such phrases often consist primarily of a verb followed by a demonstrative that functions cataphorically, signaling that further identification of an argument of the verb is to follow. The modern demonstrativemarked complement and relative constructions could easily have developed from the integration of sentences exhibiting this discourse pattern. Mohawk also contains a complement construction marked by the definite particle ne. This construction shows a different prosodic pattern, in which the particle is grouped with the following dependent clause. The prosodic difference suggests a different path of development. The ne complement constructions could not have arisen from simple integration, because clauses beginning with ne cannot function alone as independent sentences (except of course where the ne is part of a nominal argument). It might instead have developed from a construction originally marked only by prosodic integration, with subsequent extension of the definite marker from simple nominal arguments to clausal arguments. Finally, there is a restrictive relative-like construction marked by the particle né:ne or né: ‘that (is)’, common in modern Mohawk cleft constructions. Here too the marker is grouped prosodically with the following dependent clause, suggesting a different origin from the other demonstrative-marked constructions. This restrictive relative construction appears to have developed by simple prosodic integration of two sentences, the second a cleft: ‘I saw the boy. That’s the one who sleeps here.’ There is evidence that the structures seen here, those marked only prosodically and those marked by demonstratives and the article, are relatively young.

Re(e)volving complexity 

The markers have not become full-fledged complementizers: they still retain the full range of deictic functions they serve with lexical nouns. The demonstratives kí:ken ‘this’ and thí:ken ‘that’ distinguish distance in space, time, or discourse. The definite article ne marks only clauses conveying information that is generally already known to the listener. This does not of course mean that Mohawk developed complexity only recently. There is ample evidence elsewhere in the language of ancient syntactic complexity, particularly in the elaborate verb morphology. There are also several additional marked complement and relative-like constructions not described here. The young Mohawk constructions show that if we are to understand the nature of complexity and the ways in which it can emerge in a language, it is important to include consideration of prosody. A close examination of prosody allows us to perceive complexity even in the absence of overt grammatical markers, and differences in the prosodic patterns of different constructions shows that complex constructions can develop along a variety of pathways.

Abbreviations agt = grammatical agent, cisloc = cislocative, dir = directional applicative, distr = distributive, dv = duplicative, fact = factual, hab = habitual aspect, lk = linker, m = masculine gender, mid = middle, neg = negative, pat = grammatical patient, pl = plural, prf = perfective aspect, prt = partitive, sg = singular, trloc = translocative.

References Bolinger, D. 1984. Intonational signals of subordination. Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics Society 401–413. Bolinger, D. 1989. Intonation and its Uses: Melody in Grammar and Discourse. Stanford CA: Stanford University Press. Chafe, W. 1979. The flow of thought and the flow of language. In Syntax and Semantics, T. Givón, (Ed.), 159–81. New York NY: Academic Press. Chafe, W. 1982. Integration and involvement in speaking, writing and oral literature. In Spoken and Written Language: Exploring Orality and Literacy, D. Tannen (Ed.), 35–53. Norwood NJ: Ablex. Chafe, W. 1987. Cognitive constraints on information flow. In Coherence and Grounding in Discourse, R. Tomlin (Ed.), 21–51. Amsterdam: John Benjamins. Chafe, W. 1994. Discourse, Consciousness and Time: The Flow and Displacement of Conscious Experience in Speaking and Writing. Chicago IL: University of Chicago. Diessel, H. 2005. The Acquisition of Complex Sentences. Cambridge: CUP. Everett, D. 2005. Cultural constraints on grammar and cognition in Pirahã: Another look at the design features of human language. Current Anthropology 46(4): 621–646.

 Marianne Mithun Everett, D. 2007. Cultural constraints on grammar in Pirahã: A reply to Nevins, Pesetsky, and Rodrigues 2007. Ms. Givón, T. 2002. Bio-linguistics: The Santa Barbara Lectures. Amsterdam: John Benjamins. Givón, T. 2005. Context as Other Minds. Amsterdam: John Benjamins. Givón, T. (in this volume). Multiple routes to clause union: The diachrony of complex verb phrases. Hauser, M.D., Chomsky, N. & Fitch, W.T. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569–79. Heine, B. 2008. From nominal to clausal morphosyntax: complexity via expansion. International Symposium on the Rise of Syntactic Complexity. Ms. Heine, B. & Kuteva, T. 2007. The Genesis of Grammar: A Reconstruction. Oxford: OUP. Mithun, M. In press a. The fluidity of recursion and its implications. The Linguistic Review. Mithun, M. In press b. Threads in the tapestry of syntax: Complementation and Mohawk. In Proceedings of the Forty-second Meeting of the Chicago Linguistics Society, R. Peachey et al. (Eds), Chicago IL: University of Chicago Press. Mithun, M. 2007. Alternative pathways to relativization. Seminario de complejidad sintáctica. Estudios lingüistico-tipológicos y etnoculturales en lenguas indígenas y minoritarias. Maestria en Lingüistica, Universidad de Sonora, Mexico. Nevins, A., Pesetsky, D. & Rodrigues, C. 2007. Pirahã exceptionality: A reassessment. Ms, Available at http://ling.auf.net/lingBuzz. Noonan, M. 2007. Complementation. In Language Typology and Syntactic Description, 2nd Edn. T. Shopen (Ed.), 52–150. Cambridge: CUP. Parker, A.R. 2006. Evolution as a Constraint on Theories of Syntax: The Case against Minimalism. Ph.D. Dissertation, University of Edinburgh. Pawley, A. 2000. The one-clause-at-a-time hypothesis. Perspectives on Fluency, H. Riggenbach (Ed.), 163–199. Ann Arbor MI: University of Michigan. Pawley, A. & Syder, F. 1975. Sentence formulation in spontaneous speech. New Zealand Speech Therapists’ Journal 30(2): 2–11. Pinker, S. & Jackendoff, R. 2005. The faculty of language: What’s special about it? Cognition 95: 201–236. Thompson, S. 2002. ‘Object complements’ and conversation: Towards a realistic account. Studies in Language 26: 125–164. Verhagen, A. 2005. Constructions of Intersubjectivity. Oxford: OUP.

Multiple routes to clause union The diachrony of complex verb phrases* T. Givón

University of Oregon This paper investigates the diachronic pathways that lead to the rise of complex predications. It suggests that the great variety of complex predicate constructions can be traced back to two major pathways. Both pathways begin their life as paratactic verb-complement constructions (complex VPs) under separate intonation contours. Both then condense into syntactic V-complement construction under a single intonation contour. In the first type, the complement clause begins as chained (conjoined) to the main clause, and the chain then condensed into a serial verb construction. In the second type, a finite main clause and a non-finite (nominalized) object clause undergo a similar condensation. Both types can then go on to create morphologically complex lexical verbs. Both thus share the general diachronic trend of parataxis-to-syntaxis-to-lexis, albeit with somewhat different synchronic properties of both the syntactic and lexical product.

1. Introduction This paper proceeds from two main premises:

•• That complex predicates are better viewed in the broader context of syntactic complexity; that is, syntactically complex clauses and cognitively complex events.

•• That a synchronic typology, of whatever syntactic domain, can only make sense from a diachronic perspective; that is, as a typology of the diachronic pathways that gave rise to the extant synchronic types. I will begin by outlining the two main diachronic pathways through which complex verb phrases — thus eventually clause union — may arise: clause embedding (complementation) and clause chaining (conjunction). For each of these two, the syntactic properties of the resultant complex clause are in large part predictable from its diachronic source. *I have benefitted immensely from comments on earlier versions of this chapter from Bernd Heine, Keren Rice, Andy Pawley, Östen Dahl, and other participants in the 2006 Rice Symposium on Complex Predicates, and the 2008 Rice Symposium on Syntactic Complexity.

 T. Givón

Once the two main diachronic pathways to complex verb phrases have been described, I will turn to consider a number of well-known instances of complex predications. To the extent possible, I will try to determine whether these synchronic types fits within the proposed two-way diachronic typology, and to what extent, if any, the typology may need to be expanded or enriched to accommodate more types or sub-types. Along the way it will become necessary to treat one typological parameter that often intersects with the typology of complex clauses — finiteness. The term ‘complex clause’, much like ‘complex event’, begs for some definition, however cursory. Following an earlier discussion (Givón 1991a), I will suggest that a good point of departure could be to assume the following: (1) Preliminary definition of complex clauses: a. A clause, whether simple or complex, must fall under a single intonation contour. b. A complex clause must contain multiple lexical predicates. Both definitions (1a) and (1b) are to be taken as one-way conditional implications. That is, if a clause is simple, it must fall under a single intonation contour, but not vice versa (1a). So that some expressions that fall under a single intonation contour may not be considered clauses, e.g., short interjections or other predicate-less expressions.1 In the same vein, if a clause is complex, it must have multiple predicates falling under a single intonation contour, but not vice versa (1b). So that some clauses with multiple lexical predicates may not be syntactically complex, but rather morphologically or lexically complex. One should note, lastly, that the sense of clause union used here is diachronic, so that the two main diachronic pathways that yield complex verb phrases are also the two main diachronic routes to clause union.2

2. Preliminaries 2.1 Grammaticalization, co-lexicalization and clause union Clause union has been traditionally discussed almost exclusively within the context of morphological causativization, but is in fact a much broader phenomenon. Semantically first, a wide range of syntactic-semantic configurations can be the diachronic precursors to clause-union. The common denominator to all . An event clause in natural connected discourse need not contain an explicit lexical predicate, although when it doesn’t, one is most often implicit and recoverable from the context. For an extensive discussion, see Chafe (1994) and Givón (2002 ch. 3; 2008, ch. 10). . The term ‘clause union’ was used initially in the early 1970s context of Relational Grammar, in a purely synchronic sense, dependent as it was on the notion of ‘syntactic transformation’.

Multiple routes to clause union 

types of clause-union is either the grammaticalization or co-lexicalization of one of the two predicates of the two precursor clauses. In cases where clause union arises from embedding (our Type-A, below), if the main verb grammaticalizes or co-lexicalizes semantically, it most often also grammaticalizes or co-lexicalizes syntactically; that is, it becomes an affix on the erstwhile complement verb.3 In cases where a complex verb phrase arises from clause chaining (our Type-B, below), full syntactic clause-union is less common, even when it can be shown that the precursor two events have merged cognitively/semantically. Some of the more common semantic configurations that serve as precursors to clause-union of either main type are illustrated in (3) below. Their great functional diversity also illustrates the fact that the very same type of syntactically complex clause may arise through diverse adaptive-communicative motivations.

(3) Common products of clause union:

a. Causativization: (co-lexicalization) She let-go of his hand b. Tense-aspect-modal auxiliaries: (grammaticalization) He will-eat the apple They have-eaten their supper c. Evidentials: (grammaticalization) They say she’s quitting I hear she’s quitting Suppose she quits? d. Directionals: (grammaticalization)4 em tromwey sospan i-go she threw.away pot pred-go ‘She threw the saucepan away’ e. Cognate object: (grammaticalization) She made/took a left turn f. Ideophone: (co-lexicalization) It went kapow! g. Resultative verb construction: (co-lexicalization) She shot him dead. h. Passive (grammaticalization) It was discovered by the police.

. This affixation of one verb to another is sometimes called ‘predicate raising’. . Tok Pisin (Givón 1991a), here this is a serial-verb construction.

 T. Givón

i. Cleft (grammaticalization) It was Mary who discovered it. j. WH-question (grammaticalization) It was who that you saw? k. Co-verb constructions: (grammaticalization, co-lexicalization)5 liri-ma nga-ya-naggi munybaban swim-asp I-go-past other.side ‘I swam to the other side’ l. Case-marking verb serialization:6 mo fi ade ge naka I took machete cut wood ‘I cut the wood with a machete’ The reason why clause union has always appeared so conspicuous in morphological causatives (3a) is first because morphological causatives involve co-lexicalization of the two precursor verbs (‘predicate raising’). And second, because the causative main verb in such constructions is transitive, and thus takes a direct object (the manipulee). When the complement verb is also transitive, competition arises for the direct-object slot in the merged clause, between the manipulee of the main verb and the patient of the complement. For whatever reason, such competition has always got linguists’ juices flowing.7

2.2 Functional and structural dimensions of clause union 2.2.1 Event integration and clause union: The complementation scale Perhaps the best illustration of the functional and structural features of clause union, and how the two run in parallel (‘are isomorphic’), is the complementation scale found in embedding languages such as English (Givón 1980b; 2001 ch. 12). At the top of the scale one finds morphological causatives with maximal clause-union and co-lexicalization. At the bottom are direct-quote complements of utterance verbs, . Wagiman (Wilson 1999). While recognized syntactically as a co-verb construction, the semantic configuration here is that of directional, thus akin to (3d) above. . Yoruba (Givón 1975). . This competition is part of the process of merging two sets of grammatical relations in the precursor clauses into a single set in the merged complex clause. Comrie (1976) has attempted to deal with this competition via a mechanical syntactic hierarchy. Both Shibatani (1976b) and Cole (1977/1984) have shown that the competition is resolved on semantic grounds.

Multiple routes to clause union 

falling under a separate intonation contour. The transition between the two extremes is gradual both semantically and syntactically.

(4) The complementa tion scale (Givón 2001)

Semantic scale of verbs

Syntax of COMP-clause

a. She let-go of the knife b. She made him shave

CO-LEXICALIZED COMP

c. She let him go home d. She had him arrested

BARE-STEM COMP

e. She caused him to switch jobs f. She told him to leave g. She asked him to leave h. She allowed him to leave i. She wanted him to leave j. She’d like him to leave

INFINITIVE COMP

k. She’d like for him to leave

FOR-TO COMP

SUBJUNCTIVE COMP

l. She suggested that he leave m. She wished that he would leave n. She agreed that he should leave o. She knew that he had left

p. She said that he might leave later

INDIR. QUOTE COMP

q. She said: “He will leave later”

DIR. QUOTE COMP

The semantic gradation of event integration in (4) is indeed fine, but may nonetheless be subsumed under three major features: (5) Main semantic features of event integration: i. Referential integration: The sharing of referents between the two events ii. Temporal integration: Simultaneity or direct temporal adjacency of the two events iii. Spatial integration: The sharing of the same location between the two events Other widely-discussed features, such as successful (vs. unrealized) causation, intentional (vs. accidental) causation or direct (vs. indirect) causation, are relevant primarily because they imply, directly or indirectly, either co-temporality or co-spatiality of the two events (Givón 2001, ch. 12).

 T. Givón

Syntactically, the seven syntactic forms of English complements in (4) represent a fine gradation, made possible by the interaction of the following structural features: (6) Main syntactic devices that code clause-union: a. Expression of the co-referent argument: zero vs. present b. Grammatical relations: Integrated single set vs. two distinct sets c. Adjacency of the two verbs: co-lexicalization vs. separation d. Finite verb morphology: presence vs. absence on the complement verb e. Adjacency of the two clauses: presence vs. absence of a complementizer f. Intonation contours: joint vs. separate

2.2.2 Finiteness8 Even a cursory look at the complementation scale (4) and the structural devices used to affect clause-union (6) would suggest a strong involvement of finiteness — provided one recognizes finiteness as a feature of the whole clause rather than of just the verb. Thus, the most merged complement clause at the top of scale (4) has zero subjects (4a), no independent main-clause-like set of GRs (4b), co-lexicalized verbs (4c) and non-finite verb morphology (4d), all prominent features of clausal non-finiteness. The phenomenon of finiteness — and especially its converse, non-finiteness — is best illustrated in embedding, nominalizing languages, whose subordinate clauses are often fully or partially nominalized.

2.2.2.1 Finiteness and nominalization As a syntactic rather than merely morphological process, nominalization may be characterized as follows: (7) Nominalization as a syntactic process: Nominalization is the process via which a finite verbal clause — either in its entirety or only the subject-less verb phrase — is converted into a noun phrase. A verbal clause is nominalized most commonly when it occupies a prototypical nominal position/function — subject, direct object, indirect object or nominal predicate — within another clause. The syntactic structure of the NP that arises

. The discussion of finiteness here is based on Givón (2001, vol. II), mostly on various sections of chs 11 (noun phrases), 12 (verbal complements), 14 (relative clauses) and 18 (clause chaining).

Multiple routes to clause union 

through nominalization reflects, most commonly, the structure of its precursor verbal clause — minus the adjustments that come with nominalization. Within the nominalized NP, the erstwhile verb assumes the syntactic role of head noun, while other clausal constituents — subjects, objects, verbal complements or adverbs — assume the roles of various modifiers. Nominalization is thus best described as a syntactic adjustment from the finite verbal-clause prototype to the nominal (NP) prototype (Hopper & Thompson 1984; Givón 2001, ch. 2). The major components of such adjustment, at the extreme end, are:

(8) Adjustment from the prototype finite verbal clause to the prototype noun phrase: a. The verb becomes a head noun. b. The verb acquires nominal morphology. c. The verb loses tense-aspect-modal marking. d. The verb loses pronominal agreement marking. e. The subject and/or object assume genitive case-marking. f. Determiners may be added. g. Adverbs are converted into adjectives. A simple example will illustrate the general pattern emerging out of (8), contrasting the finite clause in (9a) below with its nominalized version (9b): (9) a. Finite verbal clause: She knew mathematics extensively b. Non-finite nominalized clause: Her extensive knowledge of mathematics It is of course not an accident that finiteness has been treated traditionally as a property of verbs, since four of its salient features (8a,b,c,d) pertain to the verb. But the rest of the features (8e,f,g) pertain to other constituents of the clause. Finiteness is thus an aggregate grammatical feature of clauses. And its converse, non-finiteness, is an aggregate grammatical feature of NPs derived — historically, transformationally or analogically, depending on one’s theoretical perspective — from verbal clauses.9

. Most of the syntactic relations between clauses that were taken to be synchronic ‘transformations’ in Harris (1956) and Chomsky (1957, 1965) turn out to have at least some diachronic reality. This is analogous to Chomsky & Halle’s (1968) Sound Patterns turning out to be, primarily, a recapitulation of the history of English phonology.

 T. Givón

The traditional approach also treats finiteness as a discrete, either-or feature. But since the finite prototype (or its non-finite converse) is an aggregate of many features, finiteness is at least in principle a matter of degree. This has been already seen in the complementation scale in (4). Another illustration of this gradation may be seen in (10) below: (10) Scalarity of finiteness:

Least finite

a. Her good knowledge of math (helped a lot) b. Her knowing math well (helped) c. For her to know math so well (surely helped) d. She wanted to know math well. e. Having known math well since highschool, she … f. Knowing math as well as she did, she … g. He wished that she would know math better. h. Had she studied harder, she would have known math better. i. She knew math well.

Most finite

2.2.2.2 Nominalizing vs. finite languages The broadest cross-language typological distinction in finiteness is the seeming chasm between extreme nominalizing and extreme finite languages. In the first type, all subordinate clauses, at least historically, are nominalized. Only main clauses display fully finite structure. In the second, no clause-type at all is nominalized, and all clause-types are fully finite. I will illustrate these two extreme types in order.

A. Extreme nominalizing (embedding) languages While most languages can nominalize clauses at least to some extent, some languages practice nominalization to the extreme, so that all their subordinate clauses are nominalized to some degree, and are thus non-finite. Tibeto-Burman (Watters 1998), Turkic, Carib (Gildea 1998), Quechuan (Weber 1996), Gorokan languages of the Papuan Highlands (Thurman 1978) and No. Uto-Aztecan are conspicuous examples of this type. I will illustrate this extreme type with data from Ute (Numic, Uto-Aztecan). The three most conspicuous telltale signs of clause nominalization in Ute are: •• genitive case-marking on the subject •• nominal suffix on the verb •• object case-marking on the entire clause Compare the finite verbal clause (11a) with its various nominalized counterpart (Givón 1980a, 1993a):

Multiple routes to clause union 

(11) a. Finite clause: ta’wach yoghov}ch-i pakha-qa-’u man/subj coyote-obj kill-perf-he/him ‘The man killed the coyote’ b. Nominalized clause as subject of main clause: ta’wach-i yoghov}ch-i pakha-qa-na ka-’ay-wa-t man-gen coyote-obj kill-perf-nom/subj neg-good-neg-nom

’ura-’ay be-imm

‘That the man killed the coyote is bad’ (Hist.: ‘The man’s past-killing (of) the coyote is bad’)

c. Nominalized complement of cognition verb: mamach pucucugwa-p}ga ta’wach-i woman/subj know-rem man-gen

yoghov}ch-i coyote-obj

pakha-p}ga-na-y kill-rem-nom-obj

‘The woman knew that the man (had) killed the coyote’ (Hist.: ‘The woman knew the man’s killing (of) the coyote’)

d. Object REL-clause: yoghov}ch [ta’wach-i pakha-p}ga-na] … coyote/subj [man-gen kill-rem-nom/obj] ‘the coyote that the man killed …’ (Hist.: ‘The coyote of the man’s killing ...’) e. Subject REL-clause: ta’wach [yoghov}ch-i pakha-qa-t] … man/subj [coyote-obj kill-perf-nom/subj] ‘The man who killed the coyote …’ (Hist.: The coyote-killer man …) f. Complement of modality verb: na’acich yoghov}ch-i pakha-vaa-ch girl/subj coyote-obj/gen kill-irr-nom ‘The girl wanted to kill the coyote’ (Hist.: ‘The girl wanted coyote-killing’)

'ásti'i-p}gay-'u want-rem-she

g. Complement of manipulation verb: mamach na’acich-i yoghov}ch-i pakha-vaa-ku máy-p}ga woman/subj girl-obj coyote-obj/gen kill-irr-nom/DS tell-rem ‘The woman told the girl to kill the coyote’ (Hist.: ‘the woman told the girl (to do) coyote-killing’)

 T. Givón

h. ‘If ’/‘when’-ADV clause: ta’wach-i kani-naagh y}ga-khw, … man-gen house-in enter-sub ‘When the man entered/enters the house …’ (Lit.: ‘(upon) the man’s entering the house, …)

B. Extreme finite (‘non-embedding’) languages At the other end of the typological divide one finds languages in which all clause types are finite, including, in some languages, even lexical nominalizations. Iroquois (Mithun 1991), So. Arawak and Athabaskan languages are conspicuous examples of this type. But many serial-verb languages are just as radically finite (e.g., Senufu, Niger-Congo, Carlson 1994). I will illustrate this type with data from Tolowa Athabaskan. Consider first verb complements in Tolowa, which are all finite with, tense-aspectmodality and pronominal affixes matching the prototype main-clause pattern (Bommelyn 1997; Bommelyn & Givón 1998): (12) a.

ain clause (imperf): M nn-t}-sh-’� 2s-TH-1s-observe ‘I observe you’

b. Main clause (PERF): nn-tee-s-ii-’�-’ 2s-TH-perf-1s-observe-perf ‘I observed you’ c. V-complement (implicative, IMPERF): nn-t}-sh-’� xa-sh-tQ-sri 2s-thm-1s-observe incep-1s-L-do ‘I begin to observe you’ (Lit.: ‘I begin-do I observe you’) d. V-complement (implicative, PERF-PERF): nn-tee-s-ii-’�-’ xaa-gh-íi-Q-sr� 2s-TH-perf-1s-observe-perf incep-perf-1s-L-make/perf ‘I began to observe you’ (Lit.: ‘I began-did I observed you’) e. V-complement (non-implicative, IMPERF-IMPERF): nn-t}-sh-’� ’-uu-sh-tQ-te 2s-TH-1s-observe TH-des-1s-L-want ‘I want to observe you’ (Lit.: ‘I want I observe you’)

Multiple routes to clause union 

f. V-complement (non-implicative, PERF-IMPERF): nn-t}-sh-’� ’aa-w-íi-l-te 2s-TH-1s-observe TH-des/perf-1s-L-want ‘I wanted to observe you (but maybe didn’t)’ (Lit.: ‘I wanted I observe you’) g. V-complement (non-implicative, PERF-PERF): nn-tee-s-ii-’-’� ’aa-w-íi-Q-te 2s-TH-perf-1s-observe-perf TH-des/perf-1s-L-want ‘I wanted to observe you (and I did)’ (Lit.: ‘I wanted I observed you’) While some restrictions constrain the distribution of aspectual-modal combinations in (12), complement clauses display the very same finite structure as main clauses. Relative clauses in Tolowa are just as finite, involving no subordinating morpheme but mere juxtaposition (Valenzuela 1996; L. Bommelyn, i.p.c.): (13) a. Main clause: tr’a¸¸a xe Ø-s-ii-ts’}ms woman 3s-perf-1s-kiss ‘I kissed the woman’ b. Main clause: tr’a¸¸a xe te-s-Ø-ch’a woman TH-perf-3s-leave ‘The woman left’ c. SUBJ REL-clause: tr’a¸¸a xe [Ø-s-ii-ts’}ms] te-s-ch’a woman [3s-perf-1s-kiss] TH-perf-leave ‘The woman I kissed left’ (Hist.: ‘I kissed the woman, (and) she left’) d. Main clause: tr’a¸¸a xe ch’}sne y}-s-Ø-ts’}ms woman man TR-perf-3s-kiss ‘The woman kissed the man’ e. OBJ REL-clause: ch’}sne [Tr’a¸¸a xe y}-s-Ø-ts’}ms] te-s-ch’a man [woman TR-perf-3s-kiss] TH-perf-away ‘The man the woman kissed left’ (Hist.: ‘The man, the woman kissed him, (and) he left’)

 T. Givón

Adverbial clauses are just as finite; and often the adverbial subordinator itself is historically a finite serial-verbal construction (Hennesy 1996; L. Bommelyn i.p.c.): (14) daach’}stQna-’}n’ store-to

naa-s-ee-ya, ch’a¸¸a -[n]-t’a’ mov-perf-1s-go away-rev-fly/perf

na¸¸a -[n]-n}-sh-ch’a mov-rev-perf-1s-go.away

‘After I went to the store, I came back (home)’ (Lit.: ‘I went to the store, then flying away I came back (home)’) The extreme finiteness of Tolowa syntax is underscored most conspicuously by lexical nominalizations, which display full finite structure. Only in some agent nominalizations does one find an (optional) nominalizing suffix. And if the verb is transitive, the now-extinct old antipassive (‘impersonal object’) prefix is used. Thus (Givón 2000): (15) a. ch’-u-Q-ch’ak-ne AP-con-L-pinch-nom ‘hawk’ (‘he pinches things’) b. k’waa¸’n’-ch’-uu-le’ ON-AP-con-stick ‘mosquito’ (‘he sticks something on’) In object/patient nominalization, the old de-transitive (‘D-classifier’) prefix is used: (16) a. t}-d-k’}sh TH-D-pull ‘bow’ (‘one pulls it’) b. ye’-na-y-d-tr’}sh under-mov-TR-D-wear ‘underwear’ (‘one wears it under’) c. t}-d-Q-x}t TH-D-L-gulp ‘water’ (‘one gulps it’) And in oblique nominalizations of three-argument verbs, both the antipassive and de-transitive prefixes may be used:

Multiple routes to clause union 

(17) a. m}t-ch’}-d-Q-ts’a’s WITH-AP-D-L-whip ‘whip’ (‘one whips things with it’) b. m� ’-ch’}-gh}-d-Q-t’e’sr IN-AP-perf-D-L-write ‘book’ (‘one writes things in it’) c. k’wt-gee-na-’-d-’a ON-UP-mov-AP-D-say ‘altar’ (‘one says things on it’)

3. Two diachronic routes to clause-union 3.1 Preamble In this section we will examine the two main diachronic pathways to clause-union. The first one involves the embedding of a clause into the verb phrase as a verb complement, so that both main and complement clause now fall under a single intonation contour. In this type of complex VP, the complement event is treated analogically as a nominal object of the main clause. This syntactic metaphor is not just a convenient simile, but is supported by the fact that in all languages the verbs that take embedded complements — ‘see’, ‘hear’, ‘feel’; ‘want’, ‘finish’, ‘start’; ‘make’, ‘tell’, ‘know’, ‘remember’, ‘say’ etc. — also, overwhelmingly, take nominal objects. What is more, these verbs started historically as simple transitive verbs (Heine & Kuteva 2007). On occasion one may even find hybrid transitional constructions where both nominal object and verbal complements appear in the same clause (Givón 1991a; see section 4.6 below). The second pathway involves the condensation of a clause chain into a single serial-verb clause. The resultant complex event is treated here as the syntactic analog of clausal conjunction. As noted earlier above, the first pathway leads to a much more complete clause union. In the second, the resultant complex clause, most typically with serial verbs, often displays only partial clause-union. Four typological caveats need to be noted at this point: •• Under some syntactic conditions — most conspicuously when the two (or more) verbs are adjacent and morphologically unmarked — complete clause-union may be achieved in serial-verb clauses. •• The distinction between a nominalizing vs. finite language does not always coincide 100% with the distinction between embedding vs. chaining languages, respectively. There is a substantial correlation between the two features, but it is not absolute.

 T. Givón

•• A language may be predominantly embedding or chaining, but still have some construction of the opposite type.

•• Finally, the syntactic differences between the two major types of clause union need not imply parallel semantic differences in event integration (as suggested by Pawley 1976, 1987). Rather, these syntactic differences are the synchronic typological consequences of the different diachronic pathways.

3.2 Clause-union in equi-subject (SS) configurations 3.2.1 Verb adjacency and co-lexicalization A. Embedding languages As noted earlier above, clause union in equi-subject (SS) configurations is the main diachronic venue of grammaticalized tense-aspect-modality markers (3b), directionals (3d,h), cognate object constructions(3e), ideophone clauses (3f), resultative clauses (3g) and co-verb constructions (3k). In embedding languages, complements in such a configuration are treated analogically as nominal objects of the transitive main verb. The main verb in such configuration retains the finite inflections, such as tense-aspect-modality and pronominal affixes. The complement verb is either partially or fully nominalized, exhibiting less-finite or non-finite morphology. When full clause union occurs, the grammaticalized main verb contributes all its finite marking to the colexicalized compound verb. In both VO and OV languages, SS-complementation places the complement verb directly adjacent to the main verb, thus facilitating co-lexicalization and full clauseunion. Thus, compare the VO complementation pattern of English (18a) with the OV pattern of Ute (18b): (18) Equi-subject (SS) clause-union in embedding languages: a. English (VO) S SUBJ

VP V

COMP [S] SUBJ

VP V

Mary

[Ø]

OBJ

the book

Multiple routes to clause union 

b. Ute (OV) (Givón 1980b): S

SUBJ

VP COMP [S] SUBJ

V VP

OBJ

V

Mary [Ø] po’oqwatu Mary book/ ‘Mary finished reading the book’

-kwa -

When the main verb (‘finish’) grammaticalizes as a perfect aspect, it becomes — at least initially — a finite auxiliary that remains, morpho-syntactically, the main verb of the complex two-verb clause. This is the case in English (18a). Eventually, if that auxiliary grammaticalizes more fully, it becomes a prefix on the complement verb in a VO language, or a suffix in an OV language. With cliticization, the erstwhile auxiliary now brings along all its finite morphology to the complex main verb, as is the case in Ute (18b). B. Serial-verb languages In serial-verb languages, two major factors conspire against complete clause-union. First, the precursor chained structure quite often prevents verb adjacency, scattering object nominals between verbs. One or more of the verbs in the clause may grammaticalize or co-lexicalize semantically, but it is not adjacent to another verb. As an illustration of this, compare the SS-complementation of the embedding languages in (18a,b) above with the serial-verb languages Saramaccan (VO) and Supyire (OV) in (19a,b) below: (19) Equi-subject (SS) clause-union in serial-verb languages: a. Saramaccan (VO; Byrne 1987): S

SUBJ

VP V

VP [S] OBJ

SUBJ

VP V

a bidi-wosu [Ø] he TNSthe-house ‘He finished painting the house’ (Hist.: ‘He painted the house and finished’)

 T. Givón

b. Supyire (OV; Carlson 1994): S SUBJ

VP

VP [S] SUBJ

PERF

V

PERF

VP OBJ

V

maa [0] ’ [Ø] à u-kuntunu-sEEge and (she) PERF (she) PERF her-monkey-skin ‘... and she again took her monkey-skin ...’ (Hist.: ‘... and she returned and took her monkey-skin ...’)

When ‘finish’ in (19a) and ‘return’ in (19b) grammaticalize as aspect markers, they often have no adjacent main verb to cliticize to.

3.2.2 Finiteness gradients and grammaticalization A second factor that conspires to subvert full clause-union in serial-verb languages is the lack of clear finiteness gradients among the verbs in the serial clause. As noted above, the syntactic configuration that gives rise to clause-union in embedding languages is structured by analogy with the V-OBJ configuration of the simple clause. In such constructions, the main verb retains all finite verbal features, while the complement verb is nominalized or is less-finite. When clause-union occurs in such languages, the grammaticalized main verb — now co-lexicalized with the complement verb — contributes all its finite inflections to the new complex lexical verb. As an illustration of this, consider the Spanish auxiliaries in (20) below. (20) a. se-lo-est-amos explicando dat/3s-acc/3sm-be-1p explain/part/prog ‘We are explaining it to him/her’ b. se-lo-h-an dado dat/3s-acc/3sf-have-3p give/part/perf ‘They have given it to her/him’ In serial-verb languages, often the verbs in the precursor chain are of equal finiteness. When such a chain is condensed into a single serial clause, the verbs in the condensed clause likewise are of equal finiteness. What is more, even in languages where finite verbal morphology had consolidated on a single verb in the precursor

Multiple routes to clause union 

chain — and thus on single verb in the resulting serial clause — the most-finite verb could just easily be either the one that is de-semanticized and grammaticalized, or the one that retains its lexical-semantic function. As an example consider Miskitu (OV), where the grammaticalized verb may be the finite chain-final/clause-final one, as in ‘go’ in (21a), or the non-finite chain-medial/clause-medial one, as in ‘join’ in (21b); (Hale 1991): (21) a. Baha usus-ka pali-i wa-n that buzzard-cns fly-inf go-past/3 ‘That buzzard flew away’ Hist.: ‘The buzzard flying, it went’ b. Yang nani ulta kum maki-i bangwh-i s-na 1 pl house one build-inf join-inf be-1 ‘We are building a house together’ Hist.: ‘We building a house, joining, we are’ In Akan (Benue-Kwa; Niger-Congo), the verbs in the precursor chain or the condensed serial clause may be of equal finiteness, as in (22a). Or the grammaticalized verb may be non-finite (22b); (Osam 1997): (22) a. Kofi soa-a adaka-no ko-o skuul Kofi carry-past box-the go-past school ‘Kofi carried the box to school’ b. Kofi de abaa-no hwe-e abofra-no Kofi take stick-the whip-past child-the ‘Kofi whipped the child with the stick’ The conflation of both factors — verb dispersal and lack of single locus of finite morphology — renders clause-union in serial-verb languages much more problematic, as compared to embedding languages.10

3.3 Clause-union in switch-subject (DS) configurations A. Embedding languages Switch-subject (DS) clause union involves a family of causative or resultative constructions, where the subject/agent of the complement verb is co-referent with the object/manipulee of the main verb, with main verbs such as ‘make’, ‘cause’, ‘force’, ‘tell’ or ‘let’. In embedding languages, finite marking again . Since the intervening object (causee) is highly topical, often anaphoric and thus marked as verb inflection or zero, its ‘intervention’ between the two verbs is often illusory. Verb dispersal in serial-verb languages is thus not an absolute barrier to clause-union.

 T. Givón

gravitates to the main verb, leaving the complement verb nominalized or less-finite. In an OV language, the main causative verb in DS complementation of this type winds up adjacent to the complement verb. This makes co-lexicalization and full clause-union only a matter of time — provided the main verb is high enough on the complementation scale, as is the case with the causative construction in Ute (Givón 1980a): (23)

S SUBJ

VP OBJ

COMP [S] SUBJ

VP OBJ

CAUS

V

mamach ta’wach-i [Ø] tukuavi cíira-ti-kyay-’u woman/ man/ meat/ fry---him/her ‘The woman made the man fry the meat’

The syntactic structure in (23) is probably too abstract or ‘historical’, since full clause-union and co-lexicalization leaves a complex bi-transitive verb with two objects — one the causee, the other the patient of ‘fry’. A more realistic synchronic structure is thus (Givón 1980a): S

(24) SUBJ

VP OBJ

OBJ

V

mamach ta’wach-i tukuavi cíira-ti-kyay-’u woman/ man/ meat/ fry---3sAN ‘The woman made the man fry the meat’

In a VO language such as English, it appears first that there is no automatic verb adjacency in DS-complementation, since the object of the main verb intervenes between the two verbs:

Multiple routes to clause union  S

(25) SUBJ

VP V

OBJ

COMP [S] SUBJ

VP V

Mary

made

John

[Ø] cook

OBJ

the-meat

Over time, however, a VO language can affect predicate raising and co-lexicalization in such a construction, as in Spanish: (26) María se-la-hizo comer la manzana a Juan Mary him-it-make/pret/3s eat/inf the apple dat John ‘Mary made John eat the apple’ B. Serial-verb languages Because of the dispersal of verbs among objects, serial-verb languages often fail to achieve full co-lexicalization (6c). Their objects often intersperse with their respective verbs as distinct VPs, so that several objects in the serial clause may bear the same GR — each to its own verb (6b) (Osam 1997). And finite morphology often fails to concentrate in a single verb (6d). The only structural that device verb-serializing languages use consistently to indicate clause integration is the most universal and iconic one — intonation; so that the multi-verb serial clause falls under a unified intonation contour, with neither pause nor a subordinator (6f) (see Mithun in this volume). Thus, consider the serial resultative (DS) constructions in: (27) a. Akan (VO; Osam 1997): Esi yi-i tam-no fi-i pon-no-don Esi take-past cloth-the leave-past table-the-on ‘Esi took the cloth off the table’ (Hist.: ‘Esi took the cloth and it left the table’) b. Miskitu (OV; Hale 1991): Yang truk-kum atk-ri wa-n I truck-a sell-DS/1 go-past/3 ‘I sold the truck away’ (Hist.: ‘I sold the truck and it went away’)

 T. Givón

c. Tok Pisin (VO; Givón 1991a): … em layt nau paya i-kamap … she light now fire pred-come.up ‘… She lights the fire …’ (Hist.: ‘She lights the fire and it comes up’) d. Tok Pisin (VO; Givón 1991a): … em tromwey sospan i-go … she threw.away saucepan pred-go ‘She threw the saucepan away’ (Hist.: ‘She threw the saucepan and it went away’) e. Kalam (OV; Givón 1991a): … mon d-angiy-ek yin-ip … wood take-light-past/SEQ/DS/3s burn-perf/3s ‘… She lights the wood …’ (Hist.: ‘She takes and lights the wood and it burns’) In all these examples, the object of the first verb is co-referential with the subject of the second. Often the old switch-reference morphology of the precursor chain is left intact in the serial clause, as in (27b,e) above (see also sec. 4.4. below). But, by all available syntactic tests for GRs, the semantic ‘subject’ of the second clause is a grammatical object in the serial clause. The same also applies to bona-fide causative constructions in serial-verb languages, as in: (28) a. Supyire (OV; Carlson 1994): mii à u karima à ngukuu lyi I perf him force perf chicken eat ‘I forced him to eat the chicken’ (Hist.: ‘I forced him and he ate the chicken’) b. Ijo (OV; Williamson 1965): woni u mie-ni indi die-mi we him make-asp fish share-asp ‘We made him share the fish’ (Hist.: ‘We made him and he shared the fish’) c. Ijo (OV; Williamson 1965): ari u mie mu-mi I him make go-asp ‘I chased him away’ (Hist.: ‘I chased him and he went’)

Multiple routes to clause union 

4. The transfer of finite morphology from chains to serial clauses The morpho-syntactic properties of the serial clause are often imported wholesale from its precursor clause chain. If one catches the condensation early enough in the process, the only syntactic difference between chain and serial clause is their intonational packaging — separate clausal contours vs. a single-clause contour, respectively (Mithun in this volume). In Akan clause chains, in most tense-aspects all verbs are equally finite and carry the same tense-aspect marker. This feature is transferred intact to the condensed serial clause: (29) a. Clause-chain: Araba to-o, dwow, nyen-n, kyew-e Araba buy-past yam fry-past sell-past ‘Araba bought yam, fried it and sold it’ b. Serial-V clause: Kofi yi-i tam-no fi-i pon-no-do Kofi take-past cloth-the leave-past table-the-on ‘Kofi took the cloth off the table’ (Hist.: ‘Kofi took the cloth, and it left the table’) If the clause-chaining system has chain-medial switch-reference morphology, the entire system may be transferred from the chain to the serial clause. Thus in Miskitu, the participial/infinitive suffix serves as a chain-medial SS marker, and the finite ‘past’ suffix as a chain-medial DS marker (Hale 1991): (30) a. Participle suffix as chain-medial SS marker: Yang ulta-ra dim-i kauhw-ri I house-in enter-inf/SS fall-past/1 ‘I entered the house, and fell’ ‘Entering the house, I fell’ b. Participle suffix as clause-medial SS maker: Baha usus-ka pali-i w-an That vulture-CS fly-inf/SS go-past/3 ‘The vulture flew away’ (Hist.: ‘The vulture flew, (and) it went’) c. Past suffix as chain-medial DS marker: Witin sula-kum kaik-an plap-an he deer-one see-past/3/DS run-past/3 ‘He saw the deer, (and) it ran’

 T. Givón

d. Past suffix as clause-medial DS marker: Witin sula yab-an plap-an he deer make-past/3/DS run-past/3 ‘He made the deer run’ (Hist.: ‘He caused the deer, (and) it ran’ Substantial re-analysis of the verbal morphology, between clause-chains and serial clauses, can also occur, and is indeed predictable if enough time elapses following the onset of clause-union. Thus for example, in Kalam (Papua-New Guinea) an extensive chain-medial verb morphology signals cataphoric SS vs. DS and simultaneous vs. sequential distinctions, as well as various tense-aspectmodal categories (Pawley 1966, 1976, 1987), as in (31a) below. But in most serial clauses, most of the clause-medial verbs are stripped of all such morphology, as in (31b) (Givón 1991b): (31) a. Chain-medial DS-SS morphology: … kikaruk am-nak-nin, Chicken go-IPAST/s3-SIM/DS

nuk kimb-iy, mon kamb-ak she leave-SS wood heap-the

yupiri-sap … carry-pres/3s

‘… the chicken having escaped, she leaves and carries a heap of wood …’ b. Chain final finite morphology: … mon tip-pang kom moch g-ip … wood chop-break roll crush do-perf/3s ‘… he cuts-chops-rolls-crushes the wood …’ Only in few complex serial clauses in Kalam retain the chain-medial SS/DS morphology, integrating it into the structure of the complex serial clause, as in (Givón 1991b): (32) a. SS complements of modality verbs: … nying man-ning gi-sap … water fill-irr/SS do-pres/3s ‘… she intends to fill it with water …’ (Hist.: ‘She intends and fills it with water’) b. DS complements of causative verbs: ‘… mon d-angiy-ek yin-imb … Wood take-light- RPAST/3s/SEQ/DS burn-perf/3s … ‘… she lights the wood …’ (Hist.: ‘She takes-lights the wood and it burns’)

Multiple routes to clause union 

5. Other types of complex predicates Having established the broad context for the diachronic rise of complex predications, and two major diachronic pathways through which complex clauses arise, we are now in a good position to survey some of the better-known types of so-called complex predicates, with an eye toward understanding their diachrony.

5.1 Clearly serial This type has been surveyed extensively above. What we will note briefly here is the recruitment of serial verbs for the coding of argument structure; that is, of case-marking. This pattern can be seen in many Kwa (Benue-Congo, Niger-Congo) languages, as in (Givón 1975): (33) a. iywi awa utsi iku boy took door shut ‘The boy shut the door’

(patient; Yatye)

b. mo fi ade ge naka I took machete cut wood ‘I cut the wood with the machete’

(instrument; Yoruba)

c. o fi ogbon ge igi he took cleverness cut tree ‘He clever-ly cut the tree’

(manner; Yoruba)

d. mo so fun o I say give you ‘I said to you’

(dative; Yoruba)

e. nam utom emi ni mi do work this give me ‘Do this work for me!’

(benefactive; Efik)

f. o gbara gaa ahya he ran go market ‘He ran to the market’

(allative; Igbo)

This use of serial verbs is wide-spread, and the semantics of the small group of verbs that contribute case-marking is strikingly universal: ‘take’ (PAT, INSTR, MANN), ‘give’ (DAT, BEN), ‘go’ (ALL) and ‘come’ (ABL). These are, of course, members of a slightly larger set of the usual suspects, verbs that are repeatedly implicated in various types of grammaticalization. Thus for example, the set ‘take/have’, ‘be/ stay/sit’, ‘finish’, ‘start, ‘want’, ‘go’ and ‘come’ is prominent in the grammaticalization of tense-aspect-modality.

 T. Givón

For the purpose of the discussion here, it is important to remember that the morpho-syntax of case-marking serial clauses owes its structure largely to its diachronic precursor, the clause-chain. This is particularly striking in terms of the incomplete grammaticalization of such verbs, which often retain their formal verbal properties in spite of their new grammaticalized function (Osam 1997). Consider the case-marking use of serial verbs in an embedding/nominalizing language, where the grammaticalization and/or co-lexicalization of the erstwhile verb can be more complete. Thus, for example, Ute derives all locative case-markers from historically-still-traceable precursor verbs. But these erstwhile verbs cliticize as noun suffixes, are in most cases phonologically reduced, and carry no discernible residue of verbal properties (Givón 1996): (34) De-verbal Ute post-positions: Post-position

Source verb

-va/-pa ‘at’

-paa ‘pass (through)’ (morphologically defective; old) -kwa ‘go’ (morphologically defective; old) -chugwa ‘meet (an. obj.)’ -tugwa ’go to’ -mana ‘leave’ -cawi ‘come to’ -naagha ‘enter’ -tarugwa ‘climb’ -pa’agha ‘ascend’ -tvwa ‘descend’ -rukwa ‘descend’ -yaakwi ‘descend into’ -pawi ‘descend’

-kwa ‘to’ -chux ‘to’ (an. obj.)’ -tux ‘to’ (inan. obj) -mana ‘from’ -caw ‘Toward’ -naagh ‘in’ -tarux ‘on (top)’ -pa’agh ‘on (top)’ -tvwa ‘down’ -ruk ‘under’ -yaakwi ‘down into’ -paw ‘down’

5.2 Clearly embedded In this section we will review three well-known multi-predicate constructions, suggesting that in each case their structural properties point to a reasonably clear embedding diachronic source.

5.2.1 Cognate object constructions Cognate object constructions, as they are known in English, are another type of multipredicate clause. In such constructions, a member of a relatively small group of highly de-semanticized ‘light verbs’ carries the finite verbal morphology. Such a main verb may be followed by a nominalized verb, an adjective, an adverb, or an ideophone. The group of ‘light’ verbs that partake in this construction is small and contained a

Multiple routes to clause union 

predictable selection of the usual suspects, verb that tend to grammaticalize and/or become classificatory in many languages. In contrast, the nominalized ‘heavy verbs’ that follow contain the real semantic weight of the construction, and are much more numerous. As a brief illustration, consider: (35) a. Give: give speech/talk/lecture/demonstration/performance; give a hint, give it a thought, give a kiss, give a signal, give a break, give chase, give a try, give it a shot b. Put: put an end, put some thought into, put one’s mind to, put some effort into, put a question to, put to a vote, put to flight/sleep/work/good use, put in a good word c. Make: make a decision/effort/attempt/try/decision/error/suggestion/ mistake/promise/pass; made a turn/circle/top/start; make do without, make haste, make believe, make the grade, make good time, make eyes at, make a joke, make sense d. Pay: pay attention, pay heed, pay ones respect, pay a visit e. Throw: throw a fit/party/question/suggestion/curve f. Take: take an oath/break/leap/plunge/turn/look/leak/crap/risk; take heart, take stock of, take time to, take care of, take sick, take effect, take a stand g. Have: have a feast/ball/party/cry/laugh/doubt/idea/pity; have a problem, have a second thought, have lunch, have a meeting h. Get: get busy/mad/sad/happy/wild/corny/old/young (etc.); get going, get on in years, get along with, get to the point i. Do: do justice to, do … out of, do good, do injury, do a disservice, do a service, do a favor, do a show, do a song, do without j. Go: go nuts/mad/hungry, go well with, go too far, go fifty-fifty on the deal, go dutch, go to a lot of trouble, go against the grain, go back on one’s word, go off like a rocket, go kapow!, go bang!, go k. Come: come clean/loose/true; come to an understanding, come into blows, come to a halt, come along, come to think/believe/ understand/know/realize l. Stand/stay: stand accused/guilty/tall/corrected; stand to gain/loose, stand trial, stand to reason, stand close scrutiny; stay put/healthy/alive/ active/alert/in touch m. Turn/become: turn yellow/green/white/red/blue/hostile etc. n. Utterance verbs: utter a cry/curse, say a prayer/blessing, cry uncle, sing the praise of

 T. Givón

5.2.2 Ideophone constructions An extreme case of the ‘cognate-verb’ construction is seen in So. Bantu languages, where hundreds of multi-predicate clauses may be built by combining a single ‘light’ verb — say/do — with so-called ideophones that carry a large variety of meaning, many of them manner adverbial. Many of these ideophones are etymologically related to extant lexical verb stems. Others are perhaps onomatopoeic, and many are of undetermined origin. The ‘light’ verb say/do is the only finite verb in the ideophonic construction, and the ideophones themselves carry no verbal morphology. As a brief illustration from Tswana, consider (Cole 1955): (36) a. dithupa dine ts-arobega ts-a-re kgothu kgothu stick those they-broke they-past-say ID ID ‘the sticks broke going “snap” “snap”’ b. (na) a-ntse a-re na na na (he) he-walking he-say ID ID ID ‘(he) walking very softly’ c. pula e-ne entse e-re gwaa rain it-fall on.ground it-say ID ‘the rain fell heavily’ d. ba-bo-tsaya ba-bo-re goro fafa-tse they-it-pour.out they-it-say ID on-ground ‘they poured it down on the ground’ e. logadima lono lo-gaketse lo-re lai lai lightning that it-fierce it-say ID ID ‘the lightning was fierce, flashing repeatedly’ f. mme rraagwe a-mo-tshwaara a-mo-re thusu thusu thusu father his he-him-caught he-him-say ID ID ID

kamoretlwa Stick

‘his father caught him and hit him swish swish swish with a stick’

g. yo-le a-didimala fela a-re tuu she-be she-quiet complete she-say ID ‘she said nothing, keeping very quiet’

5.2.3 Co-verb constructions Given the last two constructions, let us consider an example of the so-called ‘co-verb’ construction. In Wagiman (Australia), a small group of light verbs, 45 in all, can head complex-predicates. These verbs take the full range of finite verbal morphology, and may

Multiple routes to clause union 

also stand alone in coding simple states or events. They form a closed lexical class which includes all the usual suspects found in the serial clauses of Benue-Kwa or the cognateverb constructions of English (both italicized in (37) below). (37) The class of ‘light verbs’ in Wagiman (Wilson 1999): hit, eat, stand, come, cut, take, put, get, turn/become, burn, step on, be, become, hear, throw, spear, cry, go, bite, cook, dream, cause, name/beget, leave/go, lose, make, tell lies, have/keep, follow, sew, love oneself, fuck, chase, see, give, fear, look for, bring, tell off, sing, stay The bulk of events/states in Wagiman are coded by combining one or more nonfinite ‘co-verbs’, with at least one ‘light’ verb heading the verb phrase. Semantically, a co-verb may code an intransitive state (‘be sick’), an intransitive event (‘swell’), an intransitive motion (‘run’), a communicative act (‘talk’), a bodily function (‘yawn’), a transitive event of impact (‘kick’) or possession (‘hold’), a bi-transitive transfer event (‘pour’), an environmental phenomenon (‘thunder’), or a manner adverbial (‘quickly’). The lexical class ‘co-verb’ is, as one would expect, large and wide open. In terms of finite marking, co-verbs can take one semantically-bleached ‘aspectual’ suffix and a number of derivational suffixes. For some illustrative examples of these constructions, consider (Wilson 1999): (38) a. liri-ma nga-ya-naggi swim-asp I-go-past ‘I swam to the other side’

munybaban other.side

b. bewh-ma nga-bu-ni boran cross-asp I-hit-past river ‘I crossed the river’ c. guk-ga nga-ge-na gahan warri-buga? sleep-asp I-put-past that child-pl ‘did you put the children to sleep?’ d. ngarrmen lem du-ng hollow.log be/prfv 3s/cut-past/pfv ‘it entered the hollow log’ e. gabarn-na wek-ga quickly-asp swallow-asp ‘he swallowed it quickly’

ga-ra-n 3s-throw-past/pfv

The suggestion that the semantically-heavy ‘co-verbs’ arose as embedded complements is strengthened by their pre-light-verb position, given the incipient — or at least reconstructable — OV order in Australian language.

 T. Givón

5.3 Complex multi-stem verbal word We come now to the more difficult cases, where multiple stems are co-lexicalized into a single verbal word. Some of these constructions may be too old to allow reconstruction of their diachronic source. But in some cases the diachronic pathway may still be transparent.

5.3.1 Pre-verbal incorporation of post-positions in Rama In some languages, the incorporation of adpositions into the verb is a diachronically recent and still ongoing process, so that the governing mechanism can be still observed. One such case has been seen in Rama (Chibchan), described by Craig & Hale (1987) and Craig (1991). In this language, post-positional phrases that code various indirect objects may either follow or precede the verb. When they precede it, the object noun may be zeroed out, in contexts of either anaphoricity or, more commonly, non-referentiality or non-topicality (antipassive).11 The remaining post-position, sitting adjacent to the verb, then cliticizes as a verbal prefix. Thus consider: (39) a. ngang an-tangi Juan-ya bed they-gave John-dat ‘they gave the beds to John’ b. ngang Juan-ya an-tangi bed John-dat they-gave ‘they gave John a bed’ c. Rama ya-an-tangi Rama dat-they-gave ‘they gave (it/something) to some Rama person’ d. ngang ya-an-tangi bed dat-they-gave ‘they gave (him) a bed’ Many of the post-positions involved here turn out to have a verbal etymology, so that ultimately their incorporation may be viewed as a type of multi-predicate construction. Rama is presently an embedding OV language. But related Chibchan and Misumalpan languages show a considerable presence of serial-verb constructions (Hale 1991; Young & Givón 1990). Given the strong finiteness gradient between the main verb and the incorporated ex-verbal stem, the source of the incorporated

. A text-based functional study by Tibbitts (1995) strongly suggests the latter.

Multiple routes to clause union 

post-positions may have been pre-verbal clausal complements. But this conclusion is not absolutely firm, and the construction could have also arisen from clause-chaining.

5.3.2 Pre-verbal incorporated preposition in Romance and Germanic Pre-verbal incorporated adpositions can be found all over Germanic and Romance, where prepositions have been incorporated as verb prefixes much like in Rama. This occurred, presumably, under the same typological (SOV word-order) and functional conditions as in Rama (zeroed indirect-objects, most likely non-referential/antipassive). By way of illustration, consider the Latin-derived abstract prepositional verbs in English, all in one way or another metaphoric extensions of concrete, often spatial expressions: (40) Prepositional prefixes in Latin-derived verbs (English):

Suggested old concrete meaning

‘close’

‘hold’

‘build’

in-clude ex-clude pre-clude con-clude se-clude oc-clude

main-tain con-struct ob-tain de-struct de-tain in-struct re-tain ob-struct per-tain re-struct(ure) con-tain at-tain enter-tain abs-tain

‘carry’

‘bend’

‘pull’

‘call’

‘press’

ex-claim de-claim re-claim pro-claim ac-claim dis-claim de-claim

ex-press im-press de-press re-press com-press op-press sup-press

‘breathe’

‘form’

com-port ex-tend ex-tract in-spire re-form ex-port in-tend de-tract ex-spire in-form im-port con-tend re-tract re-spire de-form de-port dis-tend con-tract con-spire con-form re-port at-tend at-tract a(d)-spire sub-tract per-spire

‘throw’ e(x)-ject in-ject ob-ject re-ject de-ject pro-ject tra-ject(ory)

Since the original process in Latin is rather old, no firm verbal etymology for the preposition may be available, although many possible connections between prepositions and old verb stems can be suggested. Still, given that old Latin was an embedding, nominalizing OV language, the pre-verbal position of incorporated prepositions

 T. Givón

suggest that this construction may have arisen initially through an embedding pattern, in analogy with the object-verb construction. One may as well note that the same process of incorporation still goes on in English. However, in conformance with the current VO order, prepositions are incorporated post-verbally, yielding the so-called verb-particle construction: (41) Post-verbal incorporated prepositions in English a. The window broke b. The meeting broke up (early) c. Her car broke down (on the freeway) d. Her skin broke out (in a rash) e. He turned (and left) f. (So finally) he turns up (in Las Vegas) g. They turned in (for the night) h. It turned out (that she was right) i. She worked (hard) k. It worked out (just fine) l. They worked out (in the gym) m. He worked up a sweat n. They broke the furniture o. She broke up their engagement p. They broke him in (gradually) q. He broke it down (for them into small pieces) r. He turned the key s. He turned the key over (to her) t. They turned her down (for the job) u. She turned in her report v. They shut the door w. She shut him up x. They shut the plant down y. We shut them out completely z. He shut the water off. These stranded prepositions in English, while semantically part of the verb, have not yet formally co-lexicalized with the verb. For one thing, they still retain their lexical stress. For another, in many contexts they are not adjacent to the verb, so that the order variation V-OBJ-PREP vs. V-PREP-OBJ is attested and functionally significant (Chen 1986). The syntactic pattern of this incorporation probably follows established Germanic patterns (pre-verbal in the old OV Germanic dialects).

Multiple routes to clause union 

5.3.3 Incorporated objects, instruments, adverbs and verbs in No. Uto Aztecan Nominal objects, instruments and manner adverbs can incorporate into verbs. Over time, such a process may yield complex multi-stem verbal words that are on occasion also discontinuous, stranding non-lexical elements between other parts of the compound verbal stem. As an illustration of how incorporation may over time yield complex ‘bi-partite’ verbs, consider No. Paiute, (Thornes 1996; Delancey 1999a, 1999b): (42) a. ka-t}-pongosa ma-tabui-na (ma- ‘hand’) acc-poss-arrow hand-create-asp ‘(they) hand-made their arrows’ b. t}-tama-ma o-g}-pada-na (g- ‘bite’) poss-teeth-inst 3-bite-bend-asp ‘(they) bend it by biting with their teeth’ c. i-kaazi to-noyoi my-car fist-move ‘(you) push my car’

(to- ‘fist’)

d. d}-gu-hani my/asp-fire-prepare ‘(s/he) cooks for me’

(gu- ‘fire’)

e. ta-hani foot-prepare ‘herd (sheep/cattle)’

(ta- ‘foot’)

f. ku-pi-suki fire-back-warm ‘warm one’s back at the fire’

(ku- ‘fire’, pi- ‘back’)

g. pa-ko-ma-ma’i water-face-hand-wash ‘wash one’s face’

(pa- ‘water’, ko- ‘face’, ma- ‘hand’)

h. tsa-noyoi grasp-move ‘pull’

(tsa- ‘grasp’)

i. i-giki-kuba wi-ni-u my-foot-on long-step-asp ‘(s/he) stepped on my foot’

(wi- ‘long’)

j. kosso-kimaba a-tsi-kwini-ki (tsi- ‘sharp’) fire-beside ??-sharp-stand/pl-asp ‘stick (the sticks) along the fire’

 T. Givón

While many of the affixes involved are too old to determine their etymology, it is most likely that they have been derived through the incorporation of nouns, adjectives or verbs into formerly-simple verbs. In Ute, a related Numic language, the same pattern of pre-verbal incorporation is synchronically productive as, among other things, an antipassive with non-referring objects or instruments, a semantic pattern reminiscent of Rama and Latin/English, above. Thus (Givón 1980a): (43) Object-incorporation antipassive in Ute: a. Active-transitive: ta’wach kwanach-i pakha-p}ga man/subj eagle-obj/AN kill-rem ‘The man killed the eagle’ b. Antipassive: ta’wach kwana-pakha-p}ga man/subj eagle-kill-hab ‘The man killed eagles’ ‘The man did some eagle-killing’ Object incorporation is also used in Ute nominalizations, which have the same object-suppressing antipassive flavor as their English counterparts: (44) a. Agent nominalization: ta’wach kwana-pakha-mi-t ’ura-’ay man/subj eagle-kill-hab-nom be-imm ‘The man is an eagle-killer’ (⊃ He kills eagles in general) b. Action (VP) nominalization: kwana-pakha-ta ka-’ay-wa-t eagle-kill-nom neg-good-neg-nom ‘Eagle-killing is bad’ (⊃ ‘the killing of eagles in general’)

’ura-’ay be-imm

This pre-verbal incorporation pattern is also productive in Ute with semanticallyappropriate verbs, adjectives, adverbs and instruments, as in (Givón 1980a): (45) a. sak}-paghay-’way limp-walk-imm ‘s/he limp-walks’ b. mama-paghay-’way woman-walk-imm ‘he walks like a woman’

Multiple routes to clause union 

c. wii-pakha-ux-kway-’u knife-kill-asp-rem-him/her ‘s/he killed him with a knife’ d. ’át}-may-pg}a well/good-speak-rem ‘s/he spoke well, eloquently’ The antipassive object-incorporation pattern requires no invocation of pathway to complexity beyond the OV order of No. Uto-Aztecan. This pattern may have been later extended, analogically, to incorporated verbs. These are extreme nominalizing/embedding languages, and the OV, embedding pathway is thus strongly implicated. The barestem, non-finite, status of the incorporated verbs conforms to this pattern.

5.3.4 Pre-verbal incorporated ‘adverbial’ stems in Athabaskan In Athabaskan languages, the lexical verb-sense is obtained from combinations of old verb stems with ‘adverbial’ prefixes. The latter may have begun their life as verbs, but then grammaticalized as post-positions and eventually incorporated into the verbal word (Underriner 1997; Givón 2000), much like in Rama or Latin. As an illustration of many of the adverbial prefixes with a single verb-stem, consider Tolowa Athabaskan, where the oldest of these prefixes (-na- ‘motion’) can be augmented by more recent ones, many of them with clear verbal etymology (Bommelyn 1997; Givón 2000): (46) a. na-Q-da MOV-L-run b. waa-na-Q-da c. yaa-Q-da d. daa-na-Q-da e. k’wee-na-Q-da f. tr’ee-na-Q-da g. see-na-Q-da h. tee-na-Q-da i. yee-na-Q-da j. ch’aa-Q-ms k. Qee-na-y’-Q-da l. Q-ch’aa-na-Q-da m. taa-na-Q-da n. ’ee-na-Q-da o. k’w}t-na-Q-da p. ts’ee-na-Q-da q. gee-na-Q-da r. xaa-na-Q-da

‘s/he runs’

(-na ‘move around’)

‘s/e runs that-a-way’ (-wa ‘go’) ‘s/he runs through (it)’ (-ya ‘go’) ‘s/he runs into (it)’ (-da ‘sit/live’) ‘s/he is running behind (it)’ ‘s/he runs down’ ‘s/he runs up’ ‘s/he runs under water’ ‘s/he runs under (it)’ ‘s/he runs off (road)’ (-ch’a ‘leave’) ‘we-2 run together’ (-Q- reciprocal) ‘s/he runs apart’ ‘s/he runs outward’ (-ta ‘push away’) ‘s/he runs in a circle’ ‘s/he runs upon (it)’ ‘she runs out there’ ‘s/he runs away’ ‘s/he begins to run’ (-xa ‘lift up’)

 T. Givón

The diachronic pathway through which the Athabaskan incorporation pattern arose is not altogether clear. On the one hand, Athabaskan languages are extremely finite, non-nominalizing and non-embedding. Still, it is not yet clear if serial-verb clauses — the intermediate stage of the condensation in the alternative pathway — can be found in Tolowa. Since clause-chaining is a universal phenomenon across all typologies, the initial stage of this pathway is, at least in principle, always available. However, Rice (2006) has argued that the slot in which the Athabaskan ‘adverbial prefixes’ incorporate is a nominal slot, and that the incorporated ex-verbal stems have a nominal form. This suggests that in spite of their highly non-finite syntax, Athabaskan languages created these complex predicates via the embedding-nominalizing route (type A).

6. Final reflections The two major diachronic pathways that give rise to complex predications can both lead, at least potentially, to co-lexicalization, and thus to morphologically complex verbal words. The dispersal of verbs among objects in serial clauses certainly lowers the probability such an outcome in serial-verb languages. But as the Kalam data show, this is far from absolute. My own suspicion is that Kalam serial clauses with multiple bare verb stems represent a more advanced diachronic stage, where serial verbs have by and large been stripped bare of their finite morphology (Pawley in this volume). In contrast, the serial constructions in both the Miskitu and Akan are probably diachronically younger, so that much of the verbal morphology found in clause chains is still extant in serial clauses. The discussion thus far may have left the impression that only in verbserializing languages (type B) does one start the diachronic process of assembling complex VPs from a two-clause paratactic construction. This is a misleading impression, since complex clause in embedding languages in all likelihood also arise through the condensation of paratactic two-clause precursors, where in the early stage the main and the complement clause fell under separate intonation contours (Heine & Kuteva 2007). In both major pathways, thus, the process of creating complex clauses and eventually complex predicates begins with twoclause parataxis, proceeds through the intermediate stage of complex-embedded clause under a merged intonation contour, and often ends in co-lexicalization and morphologically-complex verbal words (Dahl 2004). The distribution of finite marking is a methodological heuristic tool that makes it possible to reconstruct the diachronic pathway, be it nominalizing/embedding or clause-chaining, in the absence of written records. In the final stage of condensation, that of morphologically-complex verbal words, the telltale signs of diachrony have

Multiple routes to clause union 

been largely zeroed out, so that reconstructing the diachronic pathway that led to complex verbal words is much harder. The two main diachronic pathways to complex VPs can be thus summarized schematically as in: (47) Stage:

Embedded pathway:

Clause-chain pathway:

i. paratactic source: paratactic main+COMP main+chained COMP ii. complex clause: embedded main+COMP serial-verb clause iii. complex word: complex verbal word complex verbal word As an example of the early paratactic hybrid constructions that can lead to a syntactic, embedded V-complement construction, consider V-complements in Biblical Hebrew, where the parataxis-to-syntaxis process remained endemic across a diachronic continuum spanning over 1,000 years. Two patterns may lead to such condensation, one in Early Biblical Hebrew, the other in Late Biblical Hebrew. In both the main-clause verb takes a nominal object. In the first pattern (48a), the clause-chaining source of the complement clause is transparent. In the second (48b), a paratactic RELclause structure is implicated (Givón 1991b): (48) a. va-yar’ ’elohim ’et kol 'asher ʕasa, ve-hine ṭov and saw God acc all rel made and-lo good ‘And God saw all that he had done that it was good’ (Lit.: ‘And God saw all that he had done, and lo it was good’) [Genesis, 1.31] b. ’al ti-r’u-ni she-’ani sḥarḥoret neg you-see-me rel-I swarthy/sf ‘Don’t see me that I am swarthy’ (Hist.: ‘Don’t see me, who am swarthy’) [Song of Songs, 1.6] Since pronouns cannot take restrictive REL-clauses, the only plausible source for the complement in (48b) is a paratactic, non-restrictive clause. Similar examples of condensation of V-complement construction from parataxis to syntaxis in embedding languages can be found in Germanic (Heine & Kuteva 2007), where again an earlier paratactic configuration, with the verb taking both a nominal and a clausal object and the two source clauses falling under separate intonation contour, eventually condenses into a complex clause under a single intonation contour. That is, schematically: (49) a. I want it, to do it > I want to do it b. She knew that, he did it > She knew that he did it.

 T. Givón

In Ute, an extreme nominalizing language, alternation between the syntactic and paratactic versions of verb complements can still be found synchronically, as in: (50) a. Paratactic verb complement: mamach pucucugwa-p}ga-ax, ’uru woman/subj know-rel-it that/obj

ta’wach-i man-gen

yoghov}ch-i pakha-p}ga-na-y coyote-obj kill-rel-nom-obj

‘The woman knew it, that the man (had) killed the coyote’ (Hist.: ‘The woman knew it, that the man’s killing (of) the coyote’)

b. Syntactic verb complement: mamach pucucugwa-p}ga ta’wach-i yoghov}ch-i woman/subj know-rel man-gen coyote-obj

pakha-p}ga-na-y kill-rem-nom-obj

‘The woman knew that the man (had) killed the coyote’ (Hist.: ‘The woman knew the man’s killing (of) the coyote’)

Similar alternations can be found in early child language (see Givón, in this volume). And likewise, embedded restrictive relative clauses, whether finite or nominalized, also arise from earlier paratactic, non-restrictive clauses falling under a separate intonation contour (see Givón 2008, ch. 5). Lastly, both stages of condensation along both diachronic pathways — syntacticization (embedding) and lexicalization — are driven by functional imperatives, and are thus sensitive to usage frequency. This is the real significance of our list of the usual suspects, that ubiquitous small group of verbs whose usage frequency, even before grammaticalization, is conspicuously high in all languages. These are the verbs that retain old (‘irregular’) forms long after such forms have been leveled out in the rest of the verbal lexicon (Zipf 1935). These are the verbs that often become classificatory, generic, grammaticalized, ‘light’ (de-semanticized), and thus operators on other (‘operand’) predicates. Through whatever pathway, these high-frequency verbs partake disproportionally in multi-predicate combinations that code complex events. But it is their initial lexical meaning — general, classificatory, cognitively and communicatively central — that lends them their ubiquity.

References Bommelyn, L. 1997. Prolegomena to the Grammar of Tolowa Athabaskan. MA thesis, Dept. Of Linguistics, University of Oregon. Bommelyn, L. & Givón, T. 1998. Internal reconstruction in Tolowa Athabaskan. IV Encuentro de Linguistica en el Noroeste. Hermosillo: University of Sonora.

Multiple routes to clause union 

Byrne, F. 1987. Grammatical Relations in a Radical Creole [Creole Language Library 3]. Amsterdam: John Benjamins. Carlson, R. 1994. A Grammar of Supyire. Berlin: Mouton de Gruyter. Chafe, W. 1994. Discourse, Consciousness and Time: Displacement of Conscious Experience in Speaking and Writing. Chicago IL: University of Chicago Press. Chen, P. 1986. Discourse and particle movement in English. Studies in Language 10(1). Chomsky, N. 1957. Syntactic Patterns, Mouton: The Hague. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: The MIT Press. Chomsky, N. & Halle, M. 1968. The Sound Pattern of English. Cambridge MA: The MIT Press. Cole, D. 1955. Introduction to Tswana Grammar. London: Longman, Green & Co. Cole, P. 1977/1984. The grammatical role of the causee in universal grammar. I.J.A.L. 49(2). Comrie, B. 1976. The syntax of causative constructions: Cross-language similarities and divergences. In M. Shibatani (Ed.), 1976a. Craig, C. 1991. From verb to post-position and beyond in Rama. In E. Traugott & B. Heine (Eds), 1991. Craig, C. & Hale, K. 1987. Oblique relations and reanalysis in some languages of the Americas: Language. In Chicago Conference on Native American Language and Grammatical Typology, P. Kroeber et al. (Eds), Bloomington IN: Indiana University IULC. Croft, W., Denning, K. & Kemmer, S. (Eds). 1990. Studies in Typology and Diachrony: For Joseph H. Greenberg [Typological Studies in Language 20]. Amsterdam: John Benjamins. Dahl, O. 2004. The Growth and Maintenance of Linguistic Complexity. Amsterdam: John Benjamins. Delancey, S. 1999a. Lexical prefixes and bi-partite verbs in Klamath. I.J.A.L. 65(1). Delancey, S. 1999b. Bi-partite verb stems in western North America. Ms, University of Oregon, Eugene. Gildea, S. 1998. On Reconstructing Grammar. Oxford: OUP. Gildea, S. (Ed.). 2000. Reconstructing Grammar: Grammaticalization and the Comparative Method [Typological Studies in Language 43]. Amsterdam: John Benjamins. Givón, T. 1975. Serial verbs and syntactic change: Niger-Congo. In C. Li (Ed.), 1975. Givón, T. 1980a. Ute Reference Grammar. Ignacio CO: The Ute Press. Givón, T. 1980b. The binding hierarchy and the typology of complements. Studies in Language 4(3). Givón, T. 1991a. Some substantive issues concerning verb serialization. In C. Lefebvre (Ed.), 1991 Givón, T. 1991b. The evolution of dependent clause morpho-syntax in Biblical Hebrew. In E.C. Traugott & B. Heine (Eds), 1991, Vol.2. Givón, T. 1993. Nominalized clauses in Ute: The diachronic seesaw of finite structure. II Encuentro de Linguistica en el Noroeste. Hermosillo: University of Sonora. Givón, T. 1996. La gramaticalización de verbos a posposiociones en el Ute. III Encuentro de Linguistica en el Noroeste, Hermosillo: Universidad de Sonora. Givón, T. (Ed.). 1997. Grammatical Relations: A Functionalist Perspective [Typological Studies in Language 35]. Amsterdam: John Benjamins. Givón, T. 2000. Internal reconstruction: As method, as theory. In S Gildea (Ed.), 2000. Givón, T. 2001. Syntax, Vols 1&2. Amsterdam: John Benjamins. Givón, T. 2002. Bio-linguistics. Amsterdam: John Benjamins. Givón, T. 2008. The Genesis of Syntactic Complexity: Diachrony, Ontogeny. Neuro-Cognition, Evolution. Amsterdam: John Benjamins. Givón, T. (in this volume) The ontogeny of complex verb phrases: How children learn to negotiate fact and desire.

 T. Givón Hale, K. 1991. Misumalpan verb-sequencing constructions. In C. Lefebvre (Ed.), 1991. Harris, Z. 1956. Co-occurrence and transformations in linguistic structure. Language 33: 283–340. Heine, B. & Kuteva, T. 2007. The Genesis of Grammar. Oxford: OUP. Hennesy, P. 1996. Adverbial clauses in Tolowa. Ms, University of Oregon. Hopper, P. & Thompson, S. 1984. The discourse basis for lexical categories in universal grammar. Language 60(4): 703–752. Lefebvre, C. (Ed.). 1991. Serial Verbs: Grammatical, Comparative and Cognitive Approaches, [Studies in the Sciences of Language Series 8], Amsterdam: John Benjamins. Mithun, M. 1991. The role of motivation in the emergence of grammatical categories: The grammaticalization of subjects. In E.C. Traugott & B. Heine (Eds), 1991. Mithun, M. (in this volume). Reevolving complexity: Adding intonation. Osam, E.K. 1997. Grammatical relations in Akan. In T. Givón (Ed.), 1997. Pawley, A. 1966. The Structure of Karam: A Grammar of a New Guinea Highlands Language. Ms, Aukland University. Pawley, A. 1976. On meeting a language that defies description by ordinary means, 13th Conference of the Linguistics Society of Papua-New Guinea ms Pawley, A. 1987. Encoding events in Kalam and English: Different logics for reporting events. In R. Tomlin (Ed.), 1987. Pawley, A. (in this volume). On the origins of serial verb constructions in Kalam. Rice, K. 2006. Verb stem incorporation in Athabaskan languages? The case of activity incorporates, paper given at the Rice University symposium on Complex Predicates, Houston, March 2006. Shibatani, M. (Ed.). 1976a. The Grammar of Causative Constructions, Syntax and Semantics 6. New York NY: Academic Press. Shibatani, M. 1976b. The grammar of causative construction: A prospectus. In M. Shibatani (Ed.), 1976a Thornes, T. 1996. Yahooskin–No. Paiute Verb Morphology. MA thesis, Linguistics Dept., University of Oregon. Thurman, R. 1978. Interclausal Relations in Chuave, MA thesis, UCLA. Tibbitts, B. 1995. The discourse-pragmatic context for pre-verbal postposition incorporation in Rama. Ms, University of Oregon, Eugene. Tomlin, R. (Ed.). 1987. Coherence and Grounding in Discourse [Typological Studies in Language 11]. Amsterdam: John Benjamins. Traugott, E.C. & Heine, B. (Eds). 1991. Approaches to Grammaticalization [Typological Studies in Language 19]. Amsterdam: John Benjamins. Valenzuela, P. 1996. Relative clauses in Tolowa Athabaskan. Ms, University of Oregon. Underriner, J. 1997. Notes on adverbial prexixes in Tolowa Athabaskan, Linguistics Department. Ms, University of Oregon, Eugene. Watters, D. 1998. The Kham Language of West-Central; Nepal Takale Dialect. Ph.D. dissertation, University of Oregon, Eugene. Weber, D. 1996. Una gramatica del Quechua del Huallaga [Seria Linguistica Peruana 40]. Lima: Instituto Linguístico de Verano. Williamson, K. 1965. A Grammar of the Kolokuma Dialect of Ijo [West African Language Monograph 2]. Cambridge: CUP. Wilson, S. 1999. Co-verbs and Complex Predicates in Wagiman. Stanford CA: CSLI. Young, P. & Givón, T. 1990. The puzzle of Ngabere auxiliaries: Grammatical reconstruction in Chibchan and Misumalpan, In W. Croft et al. (Eds), 1990. Zipf, G. 1935. The Psycho-Biology of Language: An Introduction to Dynamic Philology. Cambridge MA: The MIT Press (Reprinted 1965).

On the origins of serial verb constructions in Kalam Andrew Pawley

Australian National University In Kalam, a Trans New Guinea language spoken in Papua New Guinea, there are two main types of serial verb construction (SVC), showing different degrees of morphosyntactic complexity. Compact SVCs contain from two to four verb roots that form a single, semantically and syntactically very tight-knit verb phrase. Narrative SVCs depict a sequence of events that make up a familiar episode. They contain from two to five small verb phrases, compressed into a single clause– like construction. The paper will discuss the functions and origins of these two constructions and reflect on the paradox that while condensing multi-clause constructions into a single clause may simplify the task of speech planning it has created a clause type of exceptional complexity.

1. What are clauses good for?1 In 1975 I attended a mind-stretching course on language taught by George Grace at the University of Hawaii. In this course and in two later books (Grace 1981, 1987) he reminded us that human languages have evolved as devices for saying things.2 He commented that linguists in general, preoccupied with formal structures, had paid little direct attention to the question of what is entailed in saying something. This seemingly innocuous question requires us to ask about how the brain works in making sense of the world, how speakers turn perceptions and thoughts into linguistic expressions, and (among other things) how they connect these expressions with the minds of their

1. I am grateful to the Wenner-Gren Foundation, the University of Auckland and the University of Papua New Guinea for supporting my fieldwork on Kalam, in various spells between 1963 and 1993. 2. A few years later I read another mind-stretching book, On Understanding Grammar, by Tom Givón (1979), where, among other things, he explored the implications of viewing grammar as a processing strategy. While Givón argued that the structure of languages is shaped by many other forces beyond the need to describe events, he acknowledged a key role in language that is played by the clause, as a level that deals with the specification of events and states.

 Andrew Pawley

audience. In the 1970s most linguists were content to leave such concerns to philosophers and psychologists.3 The core ingredient in saying something is specifying a conceptual event or situation (Grace uses ‘event’ as a cover term for both). Following Greenberg (1959), Grace observed that only human languages can analytically specify events, saying, e.g., who did what with whom and to whom, when and where. In order to be seen as saying something, however, the speaker must also (a) contextualise the expression for the audience, e.g., by connecting it to the previous discourse or to assumed shared knowledge, (b) give it a modality (as assertion, question, negation, contingency, etc.), and (c) take responsibility for the expression (as its author or sayer, in contrast, e.g., to reading a passage from a book). The quintessential linguistic device for constructing a sketch of a single conceptual event is, of course, the independent clause or simple sentence. The prototypical clause has a single verb that represents the action or state. (For the moment I will leave aside the problematic nature of event segmentation and simply assume that each lexical verb in an utterance represents a separate conceptual event or sub-event.)4 Multi-clause constructions, by contrast, are used to construct more complex propositions. Grace’s observations set me thinking about the relation between clauses and events in Kalam, a language I had worked on for some years, and which belongs to the large (400 member) Trans New Guinea (TNG) family.5 Kalam has clause-like constructions that depart quite sharply from the one lexical verb per clause prototype. In narrative

3. Among the linguists who in the 1970s were concerned with several of these issues were Dwight Bolinger, Wallace Chafe, Tom Givón, Koenraad Kuiper, Pim Levelt and Diana Van Lancker. The field has broadened greatly in more recent times, as the discussion in Chafe (1994) and Wray (2003) indicates. 4. There is a large literature on the event structure of verbs, much of it treating English, and to a lesser extent, Romance and Germanic languages, and Hebrew. See, for example, Bohnemeyer et al. 2007; Croft 1990; Dowty 1979; Jackendoff 1990; Levin and Rappaport-Hovav 1995, 1996; Parsons 1990 and Talmy 2000. For a debate about event segmentation in Kalam see Givón (1990, 1991) and Pawley (1987, in press a). 5. Kalam is spoken by about 20,000 people living around the junction of the Bismarck and Schrader Ranges, Madang Province, Papua New Guinea. There are two main dialects, Etp and Ti, which show considerable differences in morphological forms and lexicon. Examples cited here are from the Ti dialect as spoken at Gobnem in the Upper Kaironk Valley. The main source of examples cited here is an extensive collection of tape-recordings and texts on Kalam traditional knowledge and use of animals and plants in the Ti dialect by Ian Saem Majnep and his collaborators, chiefly Majnep and Bulmer (1983, 1990, n.d.) and Majnep and Pawley (n.d.). There is a fairly extensive literature on Kalam linguistics. The most detailed study of serial verb constructions is Lane (2007), but see also Givón (1990, 1991), Pawley (2008, in press a), Pawley and Lane (1998). Other works on Kalam grammar and lexicon include Pawley (1966, 1987, 1993, in press b), Pawley et al. (2000), and Pawley and Bulmer (2003).

On the origins of serial verb constructions in Kalam 

speech it is quite common to find a serial verb construction (SVC) that contains several lexical verbs strung together under a single intonation contour with very little other material in the clause. Example (1) contains such a construction with seven verbs. (Here and in later examples, verb roots and their glosses appear in bold face. In multi-clause examples, successive clauses are distinguished as i, ii, etc. Usually each clause begins on a separate line.)6 (1) am mab pu-wk d ap agl kn-la-k. go wood hit-break.up get come ignite sleep-3pl-past A fairly literal English translation of (1) would occupy several clauses: ‘They went and gathered firewood and brought it, made a fire and slept.’ (mab has the senses ‘tree’, ‘wood’, ‘firewood’ and ‘fire’.) A free translation might say simply, ‘They gathered firewood for the night’, where the act of gathering can, in context, be understood as implying the other activities typically associated with this. In example (2), clauses (ii) and (ii) both contain eight verb roots, if we exclude the iteration of one verb, g ‘do’. (2) i. ....kayn ak ney awsek am-ub, dog the he alone go-perf.3sg ‘…theˉ(hunting)ˉdog,ˉheˉgoesˉoutˉalone,

6. The following abbreviations are used in glossing Kalam examples and in identifying sources of texts. ds dur fpkf fut imp kht kpl obj pl past perf pasthab prior presprog sg ss vp 1, 2, 3 – =

different Subject (from following verb) durative Some food plants of our Kalam forest (Majnep & Bulmer 1983) future imperative Kalam hunting traditions (Majnep & Bulmer 1990) Kalam plant lore (Majnep & Pawley. n.d.) object (case) plural remote past (yesterday or earlier) perfect (denotes present perfect, present habitual and today’s past) past habitual prior to (the event denoted by following verb) present progressive singular same subject (as following verb) verb phrase 1st, 2nd, 3rd person morpheme boundary within a phonological word clitic boundary within a phonological word

 Andrew Pawley

ii. ñn ak ognap wtsek d ap tan dˉ ap yap day the some pursuing get come ascend get come descend g g suw-p, d do bite-perf.3sg

someˉdaysˉheˉgoesˉaboutˉchasingˉallˉoverˉtheˉplaceˉandˉmakesˉkills,

iii. ñnˉ ak ognap wt-sek d ap tan dˉ ap yap day the some pursuing get come ascend get come descend g g met nη-l, do do not find-ss.prior someˉdaysˉafterˉchasingˉ(animals)ˉbackˉandˉforthˉandˉnotˉhavingˉ caughtˉany, iv. adkd katp ow-p. turning.backˉ(adv.) house come-perf.3sg heˉcomesˉbackˉhome.’

(khtˉch.19,ˉ#28)

The type of construction represented in (1) and (2ii, iii) is referred to here as a ‘narrative serial verb construction’. There is in principle no limit to the number of uninflected verbs that can occur in a narrative SVC. (In practice — if we exclude iteration of verb roots to show repetition or continuity — the limit seems to be about nine or ten.) Many languages that lack SVCs have ways of accommodating an extra verb or predicate phrase within a syntactic frame that is more or less clause-like, e.g., as secondary predicates or embedded small clauses.7 But not many languages allow four or five verbs, let alone nine or ten, in a clause. How did such constructions arise? Why would anyone want to squeeze a report specifying each of a long sequence of events into a single clause frame? Are these constructions really single clauses? In addressing these questions this paper will compare the forms and functions of four kinds of Kalam constructions that depict sequences of events — one multi-clause and three single-clause constructions. I will draw on a considerable body of published work on this language. More generally, I will ask whether the Kalam material provides evidence bearing on Givón’s (1979) proposal there are diachronic processes of syntacticisation that follow a sequence of condensation:

7. Ross (1975) referred to this integrative process as ‘clause-crunching’ and discussed a number of examples from English. A considerable literature has accumulated on various kinds of clause-crunching. Examples are Givón (1980) on types of clausal complements and various papers in Alsina et al. (1997) on types of complex predicate.

On the origins of serial verb constructions in Kalam 

discourse → syntax → morphology/lexicon → morphophonemics → zero by which loose, paratactic, “pragmatic” discourse structures develop — over time — into tight, grammaticalized syntactic structures. For each one of these processes one could prepare a balance sheet of communicative gains and communicative losses. The principles which control the balance of gain and loss here are, presumably, what we are investigating. (Givón 1979:208).

Before we examine the Kalam data something needs to be said about how to measure ‘syntactic complexity’. A standard measure is by depth of constituent structure — the more levels or intermediate nodes the greater the complexity, with particular weight given to the embedding of clause-like constituents inside phrases. However, one might also view the complexity of a particular construction in terms of how much brainwork is required to process the information in it, either as an encoder or decoder. (It needs to be kept in mind that in information theory the amount of ‘information’ carried by an item is measured by the number of possible choices facing the encoder or decoder.) Let me refer here to just a few of a number of studies that have looked at syntactic complexity in terms of mental processing costs. Discussion of an experimental study of Kalam that does the same can be found in section 3.1. Drawing on data from the Pear Stories narratives and conversational transcripts, Wallace Chafe has for many years argued that when speakers encode different sized units of information they make use of different kinds of consciousness or levels of attention (Chafe 1979, 1980, 1987, 1994). He distinguishes a ‘focus of consciousness’, a short-lived concentrated focusing of attention, from ‘peripheral’ or ‘semi-active consciousness’, where information is held in the mind but is not in focus. During a single focus of consciousness a speaker can encode a limited amount of information, typically one new ‘idea unit’ representing an event or state. The focus typically occurs in a pause of less than a half a second before a short burst of fluent speech. Such fluent bursts have a mean length of five to six words, typically fall under a single intonation contour and often correspond to a clause. Leaving aside memorised clause sequences, it seems that the simple (single verb) clause is a unit that encompasses roughly the amount of information that can easily be organised and encoded in a single planning act. A broadly similar proposal was made by Givón (1975, 1984), who argued that “the majority of sentences/clauses in connected discourse will have only one chunk — be it nominal, predicate (verb or adjective) or adverbial word/phrase — under the scope of asserted new information. All other elements in the clause will tend to be topical, background or presupposed old information” (1984:258). Du Bois (1987) put forward evidence that speakers, when encoding connected discourse, generally avoid more than one new argument per clause. Chafe suggests that severe constraints on the amount of new information that can be held in the mind in a moment of active consciousness reflect a limitation in the

 Andrew Pawley

evolution of the mammalian mind. “Our powers of remembering and imagining have far outstripped those of other creatures. But this development has failed to include any increase in the capacity of active consciousness… We are capable of thinking grand and complicated thoughts, but we can still focus our active consciousness on only very small parts of them at one time” (1994:140). By contrast, ‘scanning a center of interest’ is an extended process in which a certain range of related information held in ‘peripheral’ or ‘semi-active’ consciousness is explored and organised. It is typically associated with a break of more than a second in the speaker’s discourse flow. The linguistic outcome is often an extended sentence, made up of a sequence of discrete bursts of speech, each representing a different idea unit, strung together to describe, say, a single episode or scene. The long, paragraph-like macro-sentences one often finds in narrative speech result from the mind scanning a center of interest. Here the speaker holds several bits of information in mind at a less active level of consciousness and attempts to activate them, one by one. But speakers don’t know exactly how things are going to turn out and, not surprisingly, the results are not always completely coherent, from a grammatical or semantic standpoint. In the mid 1970s, while transcribing a corpus of conversational English speech, Frances Syder and I independently came to conclusions quite similar to Chafe’s about how clauses and multi-clause sentences are encoded (Pawley & Syder 1975, 1983, 2000). We observed that single independent clauses, of up to about eight words, are typically uttered as bursts of fluent speech, under a single intonation contour. In contrast, longer clauses and conjoined clauses are typically spoken as a series of intonation units, separated by (often short) pauses. And when speakers commit themselves to a more complex sentence frame, where two or more clauses are syntactically highly integrated, and where there are some novel (non-formulaic) lexical sequences, they often end up in a tangle. We concluded (a) that it is possible to encode the full lexical detail of short clauses in a single planning act, because this detail approximately matches the amount of information humans can activate and hold in their working memory and (b) that speakers cannot, in a single planning act, encode novel lexical combinations across independent clause boundaries. We referred to the latter limitation as ‘the one clause at a time constraint’.8 We also addressed the paradox that in order to have nativelike fluency in a language like a native one must be able to produce many multi-clause sequences as fluent

8. Although extensive research on the role of hesitation phenomena in speech production was carried out by experimental psychologists in the 1950s, 60 and early 70s (parts of it reviewed in Goldman-Eisler 1968 and Rochester 1973), Syder and I were unable to find that in that literature any proposals similar to the one-clause-at-a-time hypothesis.

On the origins of serial verb constructions in Kalam 

chunks, in apparent contradiction to the one-clause-at-a-time constraint. The explanation seems to be that multi-clause fluent sequences include large chunks that are memorised and so, in encoding tersm, do not represent new information. We concluded, for this and other reasons, that nativelike fluency in a language depends to a large degree on having memorised a large body of clause-sized constructions whose lexical content is completely or partly specified.

2.

Notes on Kalam grammar

Let us now turn to the Kalam material. This section outlines some features of Kalam grammar relevant to the discussion that follows.9

2.1 Word classes Of the major parts of speech — nouns, verbs, verb adjuncts, adverbs, adjectives and locatives — verbs and verb adjuncts are of particular relevance here. Verbs are the only part of speech to carry inflectional suffixes marking tense, aspect or mood, subject person-and-number, and anticipatory switch reference. Verb roots are a small, closed class with about 130 members. There are no morphological processes for deriving new verb stems. However, the stock of verb roots is augmented by several classes of multi-word predicates, including verb adjunct constructions and serial verb constructions. Verb adjuncts are words (either free form roots or derived words) that occur only as the partner of one verb root, or a few verb roots, with which they form a complex predicate, called a verb adjunct construction (VAC), e.g., suk ag- (laughing say) ‘to laugh’, kleηd am- (crawling go) ‘to crawl’, gadal–badal g- (higgledy-piggledy do) ‘place things higgledy-piggledy or criss-crossed’. (In these examples verb adjuncts and their literal glosses are underlined.) VACs form an open class of predicates with several hundred recorded members, often translatable by a single verb in English. In VACs the verb root serves as a classifier, marking the event as being of a certain general type. The verb adjunct specifies the subtype or specifies an associated activity to that depicted by the verb root. A VAC can occur as a predicate by itself or it can fill a verb slot in a serial verb construction.

9. In most respects Kalam’s morphological and syntactic patterns are typical of the Trans New Guinea (TNG) family. However, Kalam allows more elaborate serial verb constructions than most TNG languages.

 Andrew Pawley

2.2 Verbal clauses A verbal clause consists minimally of an inflected verb. In transitive clauses the unmarked order of major constituents is Subject Object Verb. If there is a secondary object it usually precedes the primary object, as in (3). (3) An np moni ñ-a-k? who you money give-3sg-past ‘Whoˉgaveˉyouˉmoney?’ Arguments already established in the discourse or otherwise recoverable from the speech context are usually omitted. Only one inflected verb is allowed in a clause. Verbal clauses are classified according to the kind of inflected verb that is the head or obligatory element. A clause headed by an independent verb can stand alone as a complete sentence. Independent verbs carry suffixes marking subject person-and-number and tense/aspect/mood with absolute reference (i.e., deictic reference with respect to the speech situation). A clause headed by a coordinate-dependent verb (see 2.3) cannot form a complete sentence but must occur in a coordinate relationship with an independent clause. (However, dependent clauses sometimes occur alone when the context allows missing material to be inferred.)

2.3 Chaining of coordinate-dependent (medial) clauses Coordinate-dependent verbs (often called medial verbs in descriptions of Trans New Guinea languages) are dependent on the final clause in a sentence for a full interpretation of their tense-aspect and subject reference. They carry suffixes marking subject and tense reference relative to the next verb: whether the verb has the same subject (SS) as the next verb or a different subject (DS), and whether the event denoted by the verb is prior to, simultaneous with or future to that of the following verb. However, in other respects they are coordinate with, rather than subordinate to the final verb, hence the name ‘coordinate-dependent’, used by Foley and Olson (1985). The most common suffixes marking same subject and relative tense are -l ‘SS:prior’, -lg ‘SS:simultaneous’ and -ng ‘SS-future’. The basic forms of the different subject markers are -e- ‘DS:prior’ and -knη ‘DS:simultaneous’. A coordinate-dependent verb marked for change of subject in the next verb carries a separate suffix marking the person-andnumber of its own subject, e.g., kn-na-knη (sleep-2sg-DS:simultaneous) ‘while you were sleeping (someone else did…)’. It is common for a long chain of medial (coordinate-dependent) clauses, marked for same subject and relative tense, to precede an independent clause. Sometimes such chains number more than 15 clauses. In example (4), clauses ii–ix constitute a chaining construction within the larger construction consisting of clauses i–x. A non-final intonation

On the origins of serial verb constructions in Kalam 

juncture (written here as a comma) must occur after each coordinate-dependent clause except the final one, that which immediately precedes the independent clause. Because zero anaphora is the norm for established subjects and objects it is often the case in such chains that clause after clause consists little more than an inflected verb, as in clauses v–ix.

(4) i. …ˉaps-basd=yad md-elgp-al wonˉok, grandmo.-grandfa.=my live-pasthab-3pl timeˉthat ‘…atˉtheˉtimeˉwhenˉmyˉgrandparentsˉwereˉalive,

ii. kmn=nen gos nη-l,x game=after thought perceive-ss.prior havingˉplanned toˉgoˉafterˉgameˉmammals, iii. am-l, go-ss.prior havingˉgoneˉout, iv. kmn tap nb ogok ti ti d-l, game food like those what what obtain-ss.prior havingˉgatheredˉvariousˉplantsˉforˉ(cookingˉwith)ˉgameˉmammals, v. ad ñb-l, cook eat-ss.prior havingˉcookedˉandˉeatenˉthem, vi. kn-l, sleep.ss.prior havingˉcampedˉoutˉovernight, vii. am-l, go.ss.prior havingˉgoneˉout, viii. ap-l, come-ss.prior havingˉcomeˉback, ix. g-elgp-al ak, do-pasthab-3pl topic thoseˉ(things)ˉtheyˉusedˉtoˉdo, x. mñi ag-ngab-in. now say-fut-1sg Iˉamˉnowˉgoingˉtoˉtalkˉabout.’

 Andrew Pawley

‘I’m now going to describe how, in the time of my grandparents, when people planned to hunt game mammals, they would go out and gather certain plants and cook them in stone ovens and eat them, and sleep out (in the forest), and after going out and coming back (to camp) they would do these things.’

2.4 Embedding Kalam freely allows clauses to be embedded as sentential complements, as in (5), and as relative clauses. (5) Yad [tumuk ag-p] nη-b-in 1sg thunder say-perf.3sg perceive-perf-1sg ‘I heard thunder’ (lit. approximately, ‘I heard thunder speak’. One cannot say ‘I heard thunder’, with a simple nominal as object.) Arguably, the entire sequence of clauses i–x of (4) above is the complement of the final verb agngabin ‘I will speak (about)’.

2.5 Serial verb constructions The predicate of a serial verb construction (SVC) in Kalam has as its nucleus a verb series, in which one or more bare verb roots precede an inflected verb root without any intervening conjunctions, as in (1), (2ii, iii) and (4v) above and in (6) below. (6) Am d aw-an! go get come-2sg.imp ‘Fetchˉ(it)!ˉ(lit.ˉ‘Goˉgetˉ(it)ˉandˉcome!’) All SVCs have a number of characteristics, grammatical, semantic and phonological, that support the view that they belong to a single clause. Only the final verb in the series is marked for tense/aspect/mood. This marker has scope over all the verbs in the SCV. All the verbs in the SCV share the same overt subject; this can be represented lexically only once and only the final verb in the series can carry a subject-marking suffix. Only one object NP can occur and this is shared by all transitive verbs in the SVC. Only one negator can occur. In most kinds of SVC it has scope over the whole verb series. The verb series is almost always uttered without internal pause and within a single intonation contour. The shortness of Kalam verb roots is an advantage. Verb roots are mostly monosyllabic and some consist of a single consonant. Thus the sequence of eight verb roots, wik d ap tan d ap yap g- (rub get come ascend get come descend do), meaning ‘massage s.o., rub. s.t. all over’, consists of just six syllables and takes no longer to say than excommunicated or indefatigably. Even including nominal and adverbial constituents, narrative SCVs seldom exceed 15 syllables and can comfortably be fitted into a single intonation contour.

On the origins of serial verb constructions in Kalam 

2.6 Canonical vs grammaticalised SVCs There are several types of SVCs. It is useful to make a first division between canonical and grammaticalised SVCs.10 In a canonical SVC each verb root has a lexical meaning and denotes a distinct conceptual (sub-)event in the event sequence denoted by the clause. To the extent that the events represented in a SVC are temporally discrete, their order matches the temporal order of the verbs that represent them. All the events are of roughly equal semantic importance, i.e., none are subordinate to another. In a grammaticalised SVC the final verb in the series, while it carries the TAM and subject-marking inflections, takes on a ‘grammatical’ meaning that is distinct from the meaning(s) it has when it stands alone as a lexical verb. For example, the ditransitive verb ñ ‘give, transfer, connect’ serves as a dative marker. An intransitive verb, md ‘stay, exist, live’, has been recruited as an emphatic continuative marker. At least six transitive verb roots have been recruited as emphatic completive markers. The six verbs, with some of their most common lexical senses, are ask ‘avoid, abandon, leave’, d ‘hold, get, control, stop’, ju ‘withdraw, extract’, l ‘put, become stable’, tk ‘sever, interrupt’, and yok ‘move away, displace’. Each verb in its completive function tends to co-occur with a different set of verbs from the others. A full account of the uses of grammaticalised verbs would require a separate paper. A fairly detailed treatment is given in Lane (2007). Some of the main features common to SVCs are further illustrated by each of the three clauses in (7). Clause i contains a canonical verb series. Clauses ii and iii each contains a verb series with a grammaticalised final verb (d- marks completive with reference to the actor having finished, l- marks the object as being completely affected by the action). (7) i. Ami… taw tb tk-l, mother step cut sever-ss.prior ‘Myˉmother…ˉhavingˉstampedˉonˉandˉclosedˉoffˉ(theˉentranceˉtoˉthe bandicoots’ˉburrow), ii. tug tb tk d-e-k,… holding.in.hand cut sever finish-3sg-past tookˉholdˉofˉthemˉ(oneˉbyˉone)ˉandˉclosedˉoffˉ(theˉentrance) iii. mey pak l-a-k mamd ak,… thus kill finish-3sg-past five that andˉinˉthisˉwayˉkilledˉallˉfive,ˉ…ˉ’

(khtˉch. 10, #21)

10. There are other kinds of non-canonical SVCs, which will not be discussed here. Some of these are described in Lane (2007) and Pawley (2008).

 Andrew Pawley

3. Types of canonical SVCs A fairly clear distinction can be made between two types of canonical SVC: compact and narrative.11

3.1 Compact SVCs Compact SVCs contain two or more lexical verb roots that form a single tight-knit predicate, as in (7i, ii) above and (8): (8) Kaj tb lak-eb-al. pig cut split-pres.prog.3pl ‘They are butchering pigs.’ (lit. cutting them up by splitting, i.e., making a first longitudinal cut) The verbs in a compact SVC denote sub-events that are close-spaced in time and typically connected in a causal chain. In some cases the connections people make between the constituent sub-events are probably grounded in innate perceptions of observed happenings. In other cases the connections depend on culture-specific knowledge of customary behavior. Syntactically, a compact SVC is a nuclear layer predicate in the sense of Foley and Van Valin (1984) and Foley and Olson (1985). No non-verb elements can be inserted between verb roots (other than verb adjuncts, which count as part of a verb). Compact SVCs have the ‘macro-event property’ defined by Bohnemeyer et al. (2007): temporal operators, such as tense markers and temporal adverbs, have scope over all sub-events in the construction. Compact SVCs fall into many types according to their particular semantic and grammatical makeup. Just a few types will be illustrated here (a fuller account appears in Pawley 2008). It is important to note that each type represents a productive pattern. For example, the verb series in (8), tb lak (cut split), represents a productive formula in which V1, tb ‘cut, chop’, combines with V2, a verb of result, where V2 can be, e.g., blok ‘distribute’, kluk ‘gouge, hollow out’, lak ‘split, pag ‘break,

11. The distinction between compact or narrative SVCs is similar to (though not identical to) that made between ‘component serialisation’ and ‘narrative serialisation by van Staden and Reesink (2008). Narrative SCVs very like those of Kalam appear in Kalam’s closest relative, Kobon (Davies 1981). Broadly similar constructions appear in some other New Guinea languages (e.g., Bruce 1988; Heeschen 2001; Farr 1999). This kind of SVC has variously been called condensed narrative (Heeschen 2001), narrative (van Staden & Reesink 2008), episodic (Farr 1999; Pawley 1987) and multi-scene (Lane 2007; Pawley & Lane 1998).

On the origins of serial verb constructions in Kalam 

snap’, sak ‘chip, break off a fragment’, tk ‘sever’, wk ‘break apart, shatter’, yk ‘open’, or yok ‘move away’. For each of the types exemplified in (9)–(12) below the productive patterns are defined by the accompanying notes. (9) Verb series denoting resultative or change of state events In the simplest case, resultative SVCs contain just two verbs: V1 is transitive and specifies an activity performed by an agent, usually involving forceful contact. V2 is intransitive and specifies a change of state or a movement undergone by an affected entity. The conventional meaning derived from the sequence is that the state or movement is the result of the first event. The overt subject of a resultative SVC is always the agent of V1. The logical subject of V2 is not represented.

pak cg- pak wk- pak sug- pug sug- puηl ask- puηl lak- taw pag yok- tb kluk yok-

(strike adhere) (strike shattered) (strike extinguished) (blow extinguished) (pierce open) (pierce split) (step.on broken displace) (cut gouge displace)

‘stick s.th. on, cause s.th. to adhere’ ‘knock s.th. to bits, shatter s.th’ ‘put out (a fire)’ ‘blow out (a flame)’ ‘prise s.th. open’ ‘split s.th by wedging or levering’ ‘break s.th. off by stepping on it’ ‘gouge s.th. out’

(10) Verb series denoting testing or discovering events An activity verb or verbs precedes the generic verb of perception and cognition, nη ‘perceive, be conscious, aware, see, hear, feel, smell, know, etc’. ag nη- ap nη- ay nη-

(say perceive) ‘ask, enquire, ask for, request’ (come perceive) ‘visit s.o., come and see s.o.’ (put perceive) ‘try to fit s.th., try s.th. on (e.g., clothing)’ d nη- (touch perceive) ‘feel s.th. by touching (deliberately)’ ñb nη- (consume perceive) ‘taste s.th.’ puηl nη- (pierce perceive) ‘probe, test by poking’ tag nη- (travel perceive) ‘sightsee, travel and see’ taw tag nη- (tread walk.about perceive) ‘test (ground, branch, etc.) by treading’

(11) Verb series denoting transfer/connection events A transitive verb precedes the generic verb of transfer, ñ ‘give, connect, etc.’, which denotes transfer of the referent of the affected object of V1 to the recipient of V2.

 Andrew Pawley

ag ñ- d jak ñ- d ñ- g ñ- ju ñ- ñag ñ-

tk ñ-

(say transfer) ‘tell s.th. to s.o.’ (get stand connect) ‘stand s.th. against a place’ (get transfer) ‘give s.th. personally, hand s.th. to s.o.’ (do transfer) ‘fit s.th. in position, connect to s.th.’ (withdraw transfer) ‘return s.th. to its owner, give back’ (shoot transfer) ‘fasten s.th., pass s.th. through and connect it (in sewing, buttoning)’ (write transfer) ‘write s.o. (a letter)’

(12) Verb series denoting transporting events A verb of manipulation, usually d ‘hold, handle, touch, get, have, control’ combines with one or more verbs of locomotion. d ap- d am- d am yok- d ap tan- d ap tan jak- d ap tan d ap yap-

(get come) (get go) (get go move.away) (get come ascend) (get come rise reach) (hold come ascend hold come descend)

‘bring s.th.’ ‘take s.th.’ ‘get rid of s.th, take s.th. away’ ‘bring s.th. up, fill s.th.’ ‘bring s.th. to the top, fill s.th. up’ ‘move s.th. up and down, move s.th. back and forth’

Some compact SVCs consist of a compact SVC plus another verb, or another compact SVC. For instance, the last verb series in (12) is a compact SVC that itself consists of two compact SVCs, d ap tan + d ap yap, whose order can be reversed. Why are compact SCVs so numerous? What is their role in Kalam discourse? An answer was proposed by Givón (1990, 1991), who carried out an experiment investigating the cognitive processing of different kinds of verb sequences in three languages spoken in Papua New Guinea: Kalam, Tairora, a very distant relative of Kalam spoken in the Eastern Highlands Province, and Tok Pisin, a creole whose grammar and semantics has been heavily influenced by Austronesian and Papuan languages of Melanesia. A six minute action film was shown to speakers of each language. Each subject was asked to provide two narratives describing what happened in the film, one spoken ‘online’ (during a second viewing of the film), one ‘(immediate) post-view’. Three types of constructions were compared where successive verbs have different degrees of contiguity and grammatical integration: independent clauses, coordinate-dependent clauses, and serial verb sequences. Kalam makes heavy use of SVCs, Tairora moderate use and Tok Pisin much less use. Kalam and Tairora both make extensive use of clause-chaining constructions, using coordinate-dependent verbs, but Tok Pisin does not have this type of construction. The hypothesis predicted that speakers will pause most often after an independent verb (not highly integrated with the next verb), less often after a coordinate-dependent

On the origins of serial verb constructions in Kalam 

verb (middling degree of integration) and least often after a serial verb (where the verbs are part of the same predicate phrase). The hypothesis was strongly confirmed. Kalam narrators paused between the verbs in a serial verb construction only in about 4 to 5% of cases, similar to the hesitation rate within single words. They paused much more often at boundaries between coordinatedependent clauses (about 23–32% in on-line narratives and about 48–60% post-view narratives) and consistently paused after independent clauses (81% on-line and 71% postview). Although Kalam speakers used far more SVCs than speakers of the other two languages, all three languages displayed similar overall patterns of pause probabilities, with inter-clause transitions showing a much higher rate of pausing than transitions between verbs in a SVC. Givón commented that: In terms of temporal packaging, serial-verb clauses, on the one hand, and prototypical main/finite clauses, on the other, behave as two extreme points on this scale: the former as co-lexical stems (or grammatical morphemes) within a clause; the latter as full-fledged independent clauses. However, chain-medial verbs exhibit pause probabilities and adjacency probabilities somewhere between the two extreme poles. (Givón 1990:49)

He concluded that serial verbs in Kalam and Tairora are consistently co-lexicalised (or in a minority of cases, co-grammaticalised) because they “display pause probabilities that fall within the range of lexical words” (1990:48). The main function of SVCs in Kalam, he said, is to augment the small stock of verb roots. That is to say, SVCs serve to encode conceptual events that are usually denoted by single verbs in languages with large open verb classes. I think Givón’s conclusions are valid for compact SVCs and grammaticalised SVCs. Plainly, many compact SVCs have meanings similar to English simple causative verbs and to certain kinds of phrasal verbs. Upwards of 500 compact verb series have been recorded and all are included in the dictionary of Kalam (Pawley & Bulmer 2003) on the grounds that they are standardised expressions. However, as we shall see, Givón’s conclusions do not fit well with narrative SVCs.12 A number of compact SVCs show morphological fusion in progress, where phonological reduction has blurred morpheme boundaries, e.g., in the following pairs the second form is now the conventional one: tk pag (sever break) ‘break s.t. in two’ > tpag-, pk pag (hit change.form/break) ‘bend forcefully, break or damage with a blow’ > ppag, and pk wk (hit break.up) ‘smash up, pound to pieces’ > puwk.

12. It can be argued that one function of narrative SVCs is to make up for Kalam’s paucity of verbs that represent scripted event sequences, such as ‘hunt’, ‘farm’, ‘dismantle’, ‘construct’, and ‘fetch’. However, the event structure of some SVCs is considerably more complex than any single verb in English.

 Andrew Pawley

How did compact SVCs arise? It is reasonable to argue that, long ago, they developed by clause union. But unlike narrative svcs (as we shall see), compact svcs are not readily paraphrasable by clause chaining constructions. Take the verb series tb wk (cut break.up) ‘cut to bits, chop up’. One can say: (13) Bangay tb-i,ˉ wk-p-in. pumpkin cut-ss.prior break.up-perf-1sg using a same-subject chaining construction but this means ‘Having cut (the) pumpkin, I then broke it up’. This implies a sequence of two distinct events, clearly separated in time, without a clear causal connection. The causal connection can be achieved by a two clause sequence which is marked for change of subject: (14) Bangay tb-e-n-k, wk-a-k. pumpkin cut-ds.prior-1sg-past break.up-3sg-past This translates as ‘I chopped the pumpkin and it broke up’ or ‘When I chopped the pumpkin it broke up’. However, such a two clause construction is not a natural way of expressing a direct causal connection, for the same reason as the English translations are not, perhaps because they separate the two events too sharply. Thus, it is difficult to argue that particular compact SVCs and chaining constructions are near functional equivalents in contemporary Kalam. When did compact SVCs originate? Such constructions are present in many branches of TNG and likely were present in the common ancestor of the family, which was probably spoken between six and ten millennia ago. While no one has done a systematic comparison of particular semantic patterns in compact SVCs, it is clear that there are some close correspondences across different branches of TNG, indicating that certain kinds of compact SVC constructions have been fairly stable for several millennia.

3.2 Narrative SVCs We can now take a closer look at narrative SVCs, which have a much more complex syntactic and semantic structure than compact SVCs. As their name suggests, narrative SVCs tell a short story, or parts of a story, in highly compressed form. The semantic links between events in such constructions differ from the direct causal chain and force dynamic links that characterise the event structure of many compact SVCs. The kinds of things that are mentioned in a narrative SVC, and their order, reflect conventions for telling a well-formed narrative. In a well-known paper on spoken narratives in English, Labov (1973:363) identifies the following major components of narratives:

1. 2. 3. 4. 5.

On the origins of serial verb constructions in Kalam 

Abstract. Announces the story and indicates what it is about. Orientation. Identifies the initial context, e.g., time, place and participants. Complicating action. Answers the question: What happened? Resolution. Reveals the outcome of the complicating action. Coda. Summary remark signaling that the narrative is finished.

Narratives may be complex, with two or more minimal narratives, or episodes, occurring within a larger story. A well-formed account of a single episode must at least describe the complicating action and the resolution, the other components being optional. Kalam narratives show similar functional parts to English narratives. The account may be spread over many clauses or be compressed into two or three clauses or even into a single clause, by the use of narrative SVCs.

3.2.1 Narratives reporting collecting expeditions The distinctive features of Kalam narrative SVCs may be illustrated by examining a class of narratives that are richly represented in our corpus: successful collecting expeditions, such as getting firewood, fetching water, picking fruit, gathering leafy greens, hunting for wild mammals on the ground or in trees, and collecting pandanus leaves to make mats or for thatching. Collecting expeditions represent a particular sort of purposeful activity, where there is both an immediate objective and an ultimate objective. Whether carried out by humans, nut-storing squirrels, nesting sparrows or nectar-gathering bees, successful collecting expeditions have four main stages: one or more actors (i) go forth in search of something, (ii) obtain it, (iii) carry the goods to a convenient place and (iv) process or otherwise dispose of them. A well-formed minimal report of a successful collecting expedition in Kalam reflects this pattern. The main elements in such a report can be summarised as follows. Major constituents of reports of successful collecting episodes 1 2 3 4 MOVEMENT COLLECTING TRANSPORT PROCESSING/ TO SCENE OF TO SCENE OF DISPOSAL COLLECTING PROCESSING

5 CODA

Stages 1–3 each describes elements of complicating action. Stage 4 describes the resolution, telling how the goods were processed or disposed of (e.g., cooked and eaten, preserved by smoking, stored, divided up, or traded). Occasionally there is a fifth stage, that closes off the episode by saying, e.g., that the actor(s) slept or came home. For each stage, speakers can choose to say what happened in more or less detail. Thus, some episodes are given extended treatment, while others are compressed into a few clauses or even into a single clause.

 Andrew Pawley

A narrative SVC reporting a collecting episode is defined as any SVC that contains two or more of stages 1–5. Predictably, there are no recorded cases of 1+3 and 1+4; these would be ill-formed because stages 2 and 3 describe pivotal event(s) in the complicating action and cannot be omitted from a report. But the corpus contains SVCs consisting of stages 1–2, 1–3, 1–4, 1–5, 2–3, 2–4, 3–4 and 4–5. Example (1) contains stages 1–5. Example (15) is about gathering ñepek herbs. Clause i contains the gathering stage, the transport stage, and the first event in the processing stage, cooking. However, the second event in this stage, eating, occurs in clause ii and the coda is given in iii. (15) i. ognap ksen nb tk d ap ad-l, (stages 2–4) sometimes new thus pick get come cook-ss.prior ‘…sometimesˉtheyˉwouldˉgatherˉandˉbringˉfreshˉonesˉ(ñepekˉherbs)ˉ andˉhavingˉcookedˉ(them), ii. ñb-l, eat.ss.priorˉ

(stageˉ4)

iii. kn-elgp-al. sleep-past.hab-pl andˉeatenˉ(them),ˉtheyˉwouldˉsleep.’ˉ(fpkfˉ#17)

(stageˉ5)

All narrative SVCs have a deeper constituent structure than compact SVCs. A maximal SVC reporting a collecting episode can be analyzed as containing five small verb phrases (VPs), each representing one stage in the episode. The boundary between each small VP is potentially a boundary between separate clauses in a chaining construction. Most often the verb series representing one stage of a narrative SVC is a compact SVC but more complex series sometimes occur. For example, the formulaic string d ap tan + d ap yap (get come ascend + get come descend) ‘go back and forth, go up and down’, which may occur in stage 1 or stage 3, itself consists of two compact SVCs. At the next level up, stages 2, 3 and 4 (collecting, transport and processing) form a constituent coordinate with stage 1 (movement to the collecting site) and with stage 5 phrase (the coda, usually sleeping or returning home). The verbs in stages 2–4 share the same object NP (the thing collected). They can fall under the scope of a single adverbial modifier, independently of 1. Finally the entire SVC forms a constituent, a large VP or predicate phrase, coordinate with the subject. Thus, the constituent structure of the highly recurrent lexical string in (16) is as follows: (16) [[am]vp [[[kmn pak]vp [d ap] vp]vp [ad ñb-]vp]vp]vp]vp g o game.mammal kill get come cook eat Narrative SVCs differ from compact SVCs in that the verbs need not be contiguous. Four kinds of non-verbal elements can intervene in certain positions, marking

On the origins of serial verb constructions in Kalam 

boundaries between the stages or small VPs. First, an object NP can (and often does) follow the Stage 1 verb(s) denoting movement to the scene of collecting. This can be seen in (17) and (18), as well as in (1) and (21). (17) .… am kas nb ogok tk dad ap-l,… (stages 1–3) go leaves such these pick carrying come-ss:prior ‘(they)ˉgoˉandˉpickˉsuchˉleavesˉandˉhavingˉbroughtˉthemˉback,…’’ (khtˉch.ˉ10,ˉ#113) Second, locative adjuncts can intervene. A locative adjunct to a stage 2 verb or verb series, as well as an object NP, can separate this from stage 1 material, as in (18). (18) Ney am okok kmn-nen gtag tag pak dad ap-l,… s/he go around game-after travel travel kill carrying come-ss:prior ‘Sheˉusedˉtoˉgoˉandˉwalkˉaboutˉkillingˉandˉbringingˉbackˉ gameˉmammals,…’(khtˉch. 10,ˉ#35) Alternatively, a locative adjunct to a stage 4 verb or verb series can occur after stage 3, as is the case in (19), in which the broad leaves of a spinach-like herb, bep, are gathered and put into an oven pit. (19) … mj bep tk d ap nb okyaη yok-l, … (stages 2–4) leaf spinach pick get come place below throw-ss:prior ‘… ˉhavingˉpickedˉandˉbroughtˉbepˉleavesˉandˉthrownˉ(them) ˉbelowˉ(intoˉanˉovenˉpit), … ’ˉ(khtˉch.ˉ1,ˉ#72) Thirdly, an adverbial modifier can occur between the stage 1 verb(s) and the following verbs. In such cases the scope of the modifier may be over the whole SVC or just over the verb(s) that follow the modifier. In (20) (not a collecting narrative) it is probable that the speaker intended kasek ‘quickly’ to modify only the final verb. (20) i …maj-wog ogok g ym-e-l, sweet.potato-garden these do plant-ds:prior-3pl ‘…ˉafterˉtheyˉhadˉmadeˉtheseˉsweetˉpotatoˉgardens, ii (kupyak) ap kasek ñb-e-k (rat) come quickly eat-ds:prior:3sg-past (theˉrat)ˉcameˉandˉsoonˉateˉ(there).’

(stages 3–4) (khtˉch.ˉ13,ˉ#68)

Fourthly, a negative clitic may precede the final verb in a narrative SVC. In compact SVCs the negative clitic must precede the entire verb series and it always has scope over the entire series. In narrative SVCs there are more options. First, the nonemphatic negator ma- can precede the entire verb series and have scope over it. Second, ma- can precede the final verb in the series but have scope over the whole series.

 Andrew Pawley

Third, ma-, or the emphatic negator met, can precede the final verb in the series, but have scope only over that verb, as in (2iii) above. Where do narrative SVCs fit in a typology of SVCs? There is no simple answer because there are many subtypes of narrative with subtly different characteristics. Narrative SVCs with an uninterrupted verb series, even those containing eight or ten verb roots, are almost invariably spoken under a single intonation contour. When the verb series is discontinuous short pauses are somewhat more frequent, and are most likely to occur after a stage 1 VP that is followed by a heavy locative and/or heavy object phrase. In such cases, the likely reason for the pause is that there is new information in the non-verbal constituents and the encoder has to pay close attention to these. Compare English single-verb clauses with heavy complements or modifier phrases, which often exhibit internal pauses (Chafe 1987, 1994). Some narrative SVCs qualify as nuclear layer constructions in Foley and Olson’s typology (the verbs are contiguous, and share all arguments and peripheral phrases). In other cases, the verbs are contiguous but the stage 1 VP appears to be joined to the other VPs at the core layer (it shares the subject but not the direct object), as is the case in (21) and (22). This type also counts as a single clause in terms of Foley and Olson’s criteria. In (21) a hunting episode is spread over two clauses. Stages 1–3 are represented in clause i while stage 4, cooking and eating, is represented in ii. The object of the stage 2 and 3 verbs occurs clause-initially in i, preceding the stage 1 verb, an indication that it is topicalised. (21) i. …kmn am pak dad ap-l, …game:mammal go kill carrying come-ss.prior ‘…ˉhavingˉgoneˉandˉkilledˉandˉbroughtˉgameˉmammals,

(stages 1–3)

ii. ad ñb-l katp seη ognl, … . (stage 4) cook eat-ss.prior house old:site those theyˉcookedˉandˉateˉthemˉatˉthoseˉoldˉhouseˉsites, … ’ (khtˉIntro,ˉ#8) Inˉ(22)ˉtheˉobjectˉnpˉisˉomitted,ˉhavingˉbeenˉestablishedˉearlierˉinˉtheˉnarrative. (22)ˉ Bin pataj ogok am yg pak dad ap-elgp-al … (stages 1–3) woman young these go dig kill carrying come-past.hab-3pl ‘Youngˉwomenˉusedˉtoˉgoˉandˉdigˉupˉandˉkillˉandˉbringˉbackˉ (theseˉbushˉrats)…’(khtˉch.ˉ13,ˉ#29) In a small minority of narrative SVCs, one VP appears to be joined to the rest at the peripheral layer (cases where the scope of a locative adjunct or an adverbial modifier, or a negator is restricted to just one of the VPs), and this type Foley and Olson would treat as two clauses. However, rather than dichotomising, I think it makes more sense

On the origins of serial verb constructions in Kalam 

to see different constructions as occupying different points on a continuum or scale, as being more or less like prototypical clauses.

3.3 How and why did narrative SVCs develop? There can be little doubt that narrative SVCs developed by clause union, as stripped down paraphrases of clause chains. Syntactically and semantically, the closest relatives of narrative SVCs are same subject clause-chaining constructions, in which the speaker uses a string of medial verbs to report a sequence of acts performed by the same actor. Comment has already been made on the close parallels between information packed into the separate stages, or little VPs, of a narrative SVC and information packed into separate clauses of same subject chaining constructions. Same subject (or same topic) clause chaining is extremely widespread across subgroups of TNG languages (Roberts 1997) and was presumably present in the common ancestor of the family. However, it seems that while many TNG languages have narrative SVCs of a sort, few show the degree of elaboration found in Kalam and its close relative, Kobon (Davies 1981). Presumably this elaboration was an innovation of the common ancestor of this subgroup. It seems that certain formal characteristics of the language ancestral to Kalam and Kobon provided conditions favorable to the compression of elaborate verb series into a clause-like construction. One such characteristic is that zero anaphora was then, as now, the norm for subjects and objects, allowing verb roots to be juxtaposed. A second is that the most common verb roots were short, mostly monosyllabic, and some consisted of a consonant alone, so that (as was noted earlier) a sequence of eight verb stems might occupy just six syllables. But the question remains: Why would speakers wish to cram several discrete stages of a narrative into a single clause? The cost of condensing a multi-clause construction into a single clause frame is a kind of clause construction that is considerably more complex than either typical single verb clauses or compact SVCs complexity. What is gained by such compression? There appear to be two kinds of gains, both having to do with packaging information for a fast ride. The first has to do with choices in ways of telling a story, in choosing how much detail to provide. Same actor chaining constructions are preferred when speakers want to individuate particular events in a narrative, whether merely to emphasise the temporal discreteness of the stages, or to elaborate on details. Narrative SVCs are preferred when speakers do not want to individuate the stages. In narrative SVCs individual events in the sequence are mentioned but in the most minimal way, with little or no use of what Labov (1973) calls evaluative devices — such as voice modulations, adverbial intensifiers and descriptive phrases — to add detail and drama to the bare bones of the reported actions. Of course, speakers narrating a particular episode can use a mixture of strategies, using single verb clauses for some stages and narrative SVCs for others.

 Andrew Pawley

The second gain, related to the first, is in economy of processing. The predicate phrase in a narrative SVCs is represented by speech formulas whose lexical content is stored in the long-term memory and which can be retrieved as an automatic chain. Significantly, a minority of same subject clause-chaining sequences correspond, unit for unit, to narrative SVCs: those that conform to the formula for a well-formed minimal narrative. And it is very probable that these are also the mostly frequent same subject clause chains. A measure of the rigidity of narrative SVCs is the fact that, speakers often recount a whole episode, with all its sub-events, even when the main point being made relates to just one sub-event in the sequence. This apparent transgression of the Gricean principle of economy can be seen both in narrative SVCs and in clause chaining constructions. Consider (23): (23) i. As nb-ak yg pak d ap ñb-l, (stages 2–4) small.mammal like-this dig kill get come eat-ss.prior ‘Afterˉdiggingˉupˉkillingˉbringingˉ(home)ˉandˉeatingˉthisˉkindˉofˉanimal, ii. b mnek wog ksenˉ ma-a-b-al. man next.day garden new not-go-perf-3pl menˉdon’tˉgoˉintoˉnewlyˉplantedˉgardensˉforˉtheˉnextˉfewˉdaysˉ [becauseˉtheˉactˉofˉkillingˉhasˉmadeˉthemˉrituallyˉunclean].’ Of the five sub-events represented in clause i it is only one, the act of killing this kind of animal, that makes a man ritually dangerous to crops. The other four sub-events (the mode of capture, transport, cooking and eating of the animal) are not strictly relevant to the point the narrator is making. Thus, a pragmatic English translation would simply say ‘After killing this kind of animal, men don’t enter newly planted gardens … ’ Why, in such cases, do speakers bother to mention the superfluous sub-events? There appear to be two mutually reinforcing factors at work: (i) habit: because the formula for the whole event sequence is stored in the long term memory it is just as easy to retrieve the whole sequence as to pick out the most salient sub-event(s), (ii) convention favours it. It is good style to mention all the sub-events in a routine narrative sequence. But a construction can be formulaic and still be multi-clausal. The chief advantage of using a one clause formula over a two or three clause formula to express the same information appears to be that the former takes less time and can more easily be uttered under a single intonation contour. This advantage was evidently enough for the ancestors of the Kalam and Kobon to develop narrative SVCs. It seems, then, that the clause frame is a natural target for encoders. While the norm is for clauses to contain just a single lexical verb, speakers will find ways and means of compressing frequently used multiclause expressions into a single clause

On the origins of serial verb constructions in Kalam 

construction. In this respect Kalam narrative SVCs are testament to human ingenuity and to the power of ‘chunking’. They show that, given the right phonological, semantic and syntactic preconditions, it is possible to stretch a clause structure to accommodate eight or ten verbs. But such complex expressions can be uttered as fluent units only because they are learnt as formulae, and do not need to be constructed bit by bit.

4. Conclusion: Kalam and the cycle of syntacticisation We have compared various Kalam constructions that depict event sequences, with an eye to their functional and diachronic relationships. Is there evidence here for the hypothesis that loose, paratactic structures develop into tighter syntactic structures and then into lexicalised or grammaticalised units? The Kalam material examined here did not include strictly paratactic discourse so that material does not bear on this part of the hypothesis. However, it did include same subject clause-chaining constructions, two kinds of lexical SVCs, and grammaticalised SVCs. The short answer is that there is much in the Kalam material that is consistent with the thesis that, over, time, speakers find ways and means of compressing highly recurrent multi-clause expressions into single clause expressions, that particular verb series come to have the status of lexical items and that some verbs take on grammatical functions. Jonathan Lane, whose recent book presents the most thorough account of Kalam SVCs, has neatly summed up the diachronic tendencies exhibited by these constructions. I can do no better than cite his remarks: Certain sequences of events tend to be coded in SVCs, in large part to marry speed of articulation with the requirements of Kalam discourse. Verb order is initially iconic with the order of events coded, recapitulating the most typical patterns of interclausal discourse. Once events get coded in SVCs, iconicity of another kind takes over. The close association of events coded by the verbs is reinforced by, among other things, the physical closeness of the stems themselves. Contiguity of stems, and the absence of morphological markers within SVCs, allows reinterpretation of the relation between stems… From coding independent events they move to being dependent on each other in some way. One manifestation of this is for the stems to become part of a larger lexical unit …. This has happened with resultative compounds, and with complex verbs of motion. The end point of this process is phonological fusion into a single word. Alternately, one stem can end up modifying the other. Essentially, one stem will begin to act as a grammatical marker. In Kalam SVCs, this correlates highly with SVC-final position. Hence discourse can be seen as being sucked inexorably into SVCs, and, through SVCs, into the lexicon or into the grammatical system. (Lane 2007:135)

 Andrew Pawley

References Alsina, A., Bresnan, J. & Sells, P. (Eds). 1997. Complex Predicates. Stanford CA: CSLI. Bohnemeyer, J., Enfield, N.J., Essegbey, J., Ibarretxe-Antuñano, I., Kita, S, Lüpke, F. & Ameka, F.K. 2007. Principles of event representation in language: The case of motion events. Language 83(3): 495–532 . Bruce, L. 1988. Serialization: From syntax to lexicon. Studies in Language 12(1):19–49. Chafe, W. 1979. The flow of thought and the flow of language. In Syntax and Semantics, Vol. 12. Discourse and Syntax, T. Givón (Ed.), 159–181. New York NY: Academic Press. Chafe, W. 1980. The deployment of consciousness in the production of a narrative. In The Pear Stories: Cognitive, Cultural and Linguistic Aspects of Narrative Production, W. Chafe (Ed.), Norwood NJ: Ablex. Chafe, W. 1987. Cognitive constraints on information flow. In Coherence and Grounding in Discourse, R. Tomlin (Ed.), 21–51. Amsterdam: John Benjamins. Chafe, W. 1994. Discourse, Consciousness and Time. The Flow and Displacement of Conscious Experience in Speaking and Writing. Chicago IL: University of Chicago Press. Croft, W. 1990. Possible verbs and the structure of events. In Meanings and Prototypes: Studies in Linguistic Categorization, S.L. Tsohatzidis (Ed.), 48–73. London: Routledge. Davies, J. 1981. Kobon [Lingua Descriptive Series 3]. Amsterdam: North-Holland. Dowty, D. 1979. Word Meaning and Montague Grammar. Dortrecht: Reidel. Du Bois, John. 1987. The discourse basis of ergativity. Language 63: 805–855. Farr, C. 1999. The Interface between Syntax and Discourse in Korafe, a Papuan Language of Papua New Guinea. Canberra: Pacific Linguistics. Foley, W. & Olson, M. 1985. Clausehood and verb serialization. In Grammar Inside and Outside the Clause, J. Nichols & A.C. Woodbury (Eds), 17–60. Cambridge: CUP. Foley, W. & Van Valin Jr., R.D. 1984. Functional Syntax and Universal Grammar. Cambridge: CUP. Givón, T. 1975. Focus and the scope of assertion: Some Bantu evidence. Studies in African Linguistics 6: 185–205. Givón, T. 1979. On Understanding Grammar. New York NY: Academic Press. Givón, T. 1980. The binding hierarchy and the typology of complements. Studies in Language 5: 163–193. Givón, T. 1984. Syntax: A Functional-Typological Introduction, Vol. 1. Amsterdam: John Benjamins. Givón, T. 1990. Verb serialization in Tok Pisin and Kalam: A comparative study of temporal packaging. In Melanesian Pidgin and Tok Pisin [Studies in Language Companion Series 20], J. Verhaar (Ed.). Amsterdam: John Benjamins. Givón, T. 1991. Serial verbs and event cognition in Kalam: An empirical study of cultural relativity. In Serial Verbs: Grammatical, Comparative and Universal grammar, C. Lefebvre (Ed.), 137–184. Amsterdam: John Benjamins. Goldman-Eisler, F. 1968. Psycholinguistics: Experiments in Spontaneous Speech. New York NY: Academic Press. Grace, G. 1975–84. Ethnolinguistic notes. Mimeo. Dept of Linguistics, University of Hawaii. Grace, G. 1981. An Essay on Language. Columbia SC: Hornbeam Press. Grace, G. 1987. The Linguistic Construction of Reality. London: Croom Helm. Greenberg, J.P. 1959. Language and anthropology. In Evolution and Language. A Centennial Appraisal, B.J. Meggers (Ed.), Washington DC: Anthropological Society of Washington.

On the origins of serial verb constructions in Kalam 

Heeschen, V. 2001. Event-formulas: Sentences as minimal narratives. In The Boy from Bundaberg: Studies in Melanesian Linguistics in Honour of Tom Dutton, A. Pawley, M. Ross & D. Tryon (Eds), 155–173. Canberra: Pacific Linguistics. Jackendoff, R. 1990. Semantic Structures. Cambridge MA: The MIT Press. Labov, W. 1973. Language in the Inner City [Ch. 9, The transformation of experience in narrative syntax.] Philadelphia PA: University of Pennsylvania Press. Lane, J. 2007. Kalam Serial Verb Constructions. Canberra: Pacific Linguistics. Levelt, W.J.M. 1989. Speaking: From Intention to Articulation. Cambridge MA: The MIT Press. Levin, B. & Rappaport Hovav, M. 1995. Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge MA: The MIT Press. Levin, B. & Rappaport Hovav, M. 1996. From lexical semantics to argument realization. In Handbook of Morphosyntax and Argument Structure. H. Borer (Ed.), Dordrecht: Kluwer. Majnep, I.S. & Bulmer, R. 1983. Some Food Plants in our Kalam Forest. [Department of Anthropology Working Paper 63]. Auckland: University of Auckland. Majnep, I.S. & Bulmer, R. 1990. Aps Basd Skop Kmn ak Pak ñbelgpal. Kalam Hunting Traditions, Vols 1–6 [Department of Anthropology Working Papers 85–90], A. Pawley (Ed.), Auckland: University of Auckland. Majnep, I.S. & Bulmer. n.d. Kalam hunting traditions, Vols 7–12, A. Pawley (Ed.), (Printout. [1990] Department of Linguistics, Research School of Pacific and Asian Studies, Australian National University). Majnep, I.S. & Pawley, A. n.d. Kalam plant lore. Computer printout. [2007] Department of Linguistics, Research School of Pacific and Asian Studies, Australian National University. Parsons, T. 1990. Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge MA: The MIT Press. Pawley, A. 1987. Encoding events in Kalam and English: different logics for reporting experience. In Coherence and Grounding in Discourse [Typological Studies in Language 11], R. Tomlin (Ed.), 329–360. Amsterdam: John Benjamins. Pawley, A. 1966. The structure of Karam: a grammar of a New Guinea highlands language. Ph.D. dissertation, Department of Anthropology, University of Auckland. Pawley, A. 1993. A language which defies description by ordinary means. In The Role of Theory in Language Description, W. Foley (Ed.), 87–129. Berlin: Mouton de Gruyter. Pawley, A. 2008. Compact versus narrative serial verb constructions in Kalam. In Serial Verb constructions in Austronesian and Papuan Languages, G. Senft (Ed.), 171–202. Canberra: Pacific Linguistics. Pawley, A. In press a. Event representation in serial verb constructions. In Event Representations in Language and Cognition, J. Bohnemeyer & E. Pederson (Eds), Cambridge: CUP. Pawley, Andrew. In press b. On the argument structure of complex predicates in Kalam. BLS S32. Pawley, A. & Bulmer, R. 2003. A dictionary of Kalam with ethnographic notes. Printout. Department of Linguistics, Research School of Pacific and Asian Studies, Australian National University. Pawley, A. & Lane, J. 1998. From event sequence to grammar: Serial verb constructions in Kalam. In Case, Typology and Grammar, A. Siewierska & J.J. Song (Eds), 201–227. Amsterdam: John Benjamins. Pawley, A. & Syder, F. 1975. Sentence formulation in spontaneous speech. New Zealand Speech Therapists’ Journal 30(2): 2–11. Pawley, A. & Syder, F. 1983. Two puzzles for linguistic theory: Nativelike selection and nativelike fluency. In Language and Communication, J.C. Richards & R.W. Schmidt (Eds), 191–227. London: Longman.

 Andrew Pawley Pawley, A. & Syder, F. 2000. The one clause at a time hypothesis. In Perspectives on Fluency, H. Riggenbach (Ed.), 163–199. Ann Arbor MI: University of Michigan Press. Roberts, J. 1997. Switch-reference in Papua New Guinea: A preliminary survey. In Papers in Papuan Descriptive Linguistics, A. Pawley (Ed.), 101–241. Canberra: Pacific Linguistics. Rochester, S.R. 1973. The significance of pauses in spontaneous speech. J. Psycholinguistic Research 2(1): 51–81. Ross, J.R. 1975. Parallels in phonological and semantactic organization. In The Role of Speech in Language, J.F. Kavanagh & J.E. Cutting (Eds), 283–304. Cambridge MA: The MIT Press. Talmy, L. 2000. Towards a Cognitive Semantics, 2 Vols. Cambridge MA: The MIT Press. van Staden, M. & Reesink, G. 2008. A functional approach to verb serialization. In Serial Verb constructions in Austronesian and Papuan Languages, G. Senft (Ed.), 17–54. Canberra: Pacific Linguistics. Wray, A. 2003. Formulaic Language and the Lexicon. Cambridge: CUP.

A quantitative approach to the development of complex predicates The case of Swedish Pseudo-Coordination with sitta “sit”* Martin Hilpert & Christian Koops

Freiburg Institute for Advanced Studies/Rice University This paper traces the historical development of the Swedish Pseudo-Coordination construction with the posture verb sitta “sit”. In Swedish a small number of verbs, including posture verbs such as sitta, are used in coordination with another verb to convey that the described event has an extended duration or is in progress. Quantitative evidence from Swedish historical corpora suggests that the construction has, even after it established itself as a grammatical construction, undergone a number of gradual changes in the course of the past five centuries. As part of the Pseudo-Coordination construction, the verb sitta has changed its argument structure, and the entire construction has increased in syntactic cohesion.

1. Introduction Complex predicates, as defined by Alsina et al. (1997) and Bowern (2008), are not a prominent feature of the Germanic languages. Studies of complex predicates in African, Asian, Australian, and Oceanic languages have addressed the phenomenon in discussions of serial verb constructions (Foley & Olson 1985; Aikhenvald 2006), light verb constructions (Butt 1995; Harris 2008, Hook & Pardeshi 2006), and restructuring predicates (Di Sciullo & Rosen 1990; Alsina 1997). These studies have provided a rich typology of the possible forms and functions. One might ask, then, what additional insights into complex predicates can be gained by studying the well-researched grammatical systems of, say, English or Swedish.

*We would like to thank Claire Bowern and the participants of the 11th Biennial Rice Linguistics Symposium for their helpful suggestions. Martin Hilpert thankfully acknowledges funding from the DAAD (German Academic Exchange Service). This article appeared originally in Diachronica 25(2), 2008.

 Martin Hilpert & Christian Koops

In this paper we argue that it is specifically the diachronic study of complex predicates that stands to benefit from a focus on Germanic. What makes this possible are the substantial amounts of historical text that are available today for most Germanic languages in electronic form. Together with appropriate corpus linguistic techniques, historical corpora allow the diachronic study of the Germanic languages at a level of detail that is not available for most of the languages treated in the works cited above (see also the comments by Butt & Lahiri 1998: 27). Our diachronic perspective thus complements the existing body of synchronic, fieldwork-based case studies. Of particular interest in this regard are, of course, typological generalizations about the historical development of complex predicates. For example, synchronic typologies of complex predicates indicate that the diachronic sources of light verbs are strikingly uniform across unrelated languages. This suggests that certain types of light verb constructions develop in quasi-universal ways. A case study of a complex predicate in a Germanic language can serve as a test case to explore this possibility, given the availability of relatively large amounts of textual data spanning several centuries. Historical corpus data have typically been used to demonstrate qualitative contrasts between examples from different time periods. To take a well-known example, consider the following examples of the English subordinating conjunction while in Old English and Early Modern English (Hopper & Traugott 2003: 85). (1) Đæt lastede þa [xix] winttre wile Stephne was king “That lasted those 19 winters while Stephen was king.” (ChronE [Plummer] 1137.36) (2)

Whill others aime at greatnes boght with blod, Not to bee great thou strives, bot to bee good. “While others aim at greatness that is bought with blood, You strive not to be great but good.” (1617, Sir W. Mure, Misc. Poems xxi.23 [OED while 2b])

The examples illustrate that the co-temporal meaning of while, which is clearly present in (1), gave rise to the concessive meaning seen in (2). The latter meaning can also be observed today in purely concessive usage, e.g., While dolphins inhabit the sea, they are actually mammals. Examples like (1) and (2) show that a change has occurred. However, the comparison of individual examples does not yield a precise picture of when this change was initiated and how it proceeded. As we see it, an important benefit of large historical corpora is the possibility of precisely quantifying shifting usage patterns of linguistic forms, especially their rate of occurrence in different grammatical contexts. Of course, for historical corpus studies to be feasible, questions about the development of complex predicates have to be phrased in such a way that quantitative evidence yields meaningful answers. In other words, it is not enough to report

A quantitative approach to the development of complex predicates 

frequencies as an end in itself. We have to develop with hypotheses that can be either corroborated or falsified through frequency data. The study of frequency patterns in language use is not universally accepted as a promising avenue of research. Bybee & Hopper (2001:1) explain this with reference to “the widespread acceptance of the premise that language structure is independent of language use”. However, a growing body of literature documents counterevidence to this premise. This evidence comes in the form of frequency effects, i.e., structural differences between related language forms which can be shown to correlate with a difference in discourse frequency. For example, high token frequency commonly leads to phonetic and phonological reduction. This explains why memory and family tend to be produced as bi-syllabic words in American English, while the phonetically similar but less frequent items mammary and homily are tri-syllabic (Hooper 1976). Another frequency effect can be observed in paradigmatic regularization. The irregular past tense forms of verbs such as weep and leap tend to be regularized to weeped and leaped, whereas the corresponding forms of the more frequent verbs keep and sleep are retained as kept and slept (Bybee 1985). At the level of syntax, recent work by Gahl & Garnsey (2004) shows that frequencies of syntactic patterns correlate with the phonetic reduction of certain elements. The authors demonstrate experimentally that past tense forms of verbs show a higher rate of final /t/ or /d/ deletion if the verb is produced in the context of a complementation pattern that is highly typical for that verb. To illustrate, a reduced production of the past tense form confirmed is more likely in (3), where the verb is followed by a direct object, than in (4), where it has a sentential complement.

(3) The CIA director confirmed [the rumor]DO once it had spread widely.

(4) The CIA director confirmed [the rumor should have been stopped sooner.]SC Gahl & Garnsey (2004:762)

The authors show that the bias of confirm to occur with direct objects, rather than sentential complements, contributes significantly to this effect. Our approach to the development of complex predicates in this paper is fully aligned with the idea that language structure is shaped by language use. More specifically, we assume that the diachronic study of frequency patterns can further our understanding of different aspects of complex predicates, such as the integration of parataxis into hypotaxis (Givón, in this volume) or the semantic development of lexical verbs into grammatical elements (Bybee et al. 1994; Kuteva 2001). Many of these issues have already been addressed on the basis of crosslinguistic synchronic data (Heine et al. 1991; Bybee et al. 1994; Svorou 1994, inter alia), and various semantic regularities across genetically unrelated languages have been established (e.g., Heine & Kuteva 2002). For instance, verbs of movement typically come to express the intentions of human agents and ultimately develop into markers of futurity, as seen for example in French aller “go” or English be going to (Bybee et al. 1994).

 Martin Hilpert & Christian Koops

Given these regular tendencies, it is possible to derive predictions about parallel diachronic developments in different languages.

2. Swedish Pseudo-Coordination If something can be learned from the quantitative, diachronic analysis of a Germanic complex predicate, what would be a good candidate for a case study? In this paper we focus on the development of the Swedish Pseudo-Coordination construction (SPC) with the posture verb sitta “sit”. Crosslinguistically, posture verbs often grammaticalize into markers of imperfective aspect (Bybee et al. 1994; Austin 1998; Kuteva 2001), so Swedish sitta can be hypothesized to have developed along a well-known grammaticalization path. A modern example of this construction is given in (5).1 (5) Vi bara satt och pratade. we just sat and talked “We were just talking.”

(New novels)

In Swedish, as in Danish and Norwegian, a small number of verbs, including posture verbs such as sitta “sit”, can be used in coordination with another verb to impose a particular aspectual contour on the event coded by the second verb. The resulting interpretation is typically that the described event has an extended duration or is in progress, i.e., as durative or progressive. This can be seen in the English translation of example (5). The translation also shows that the lexical meaning of sitta, i.e., the idea of being in a sitting posture, is not necessarily a prominent part of the sentence’s meaning (although an interpretation with an emphasis on the postural semantics of sitta is not ruled out; see below). A range of additional syntactic and semantic effects have been pointed out which distinguish the SPC from canonical coordination (Wiklund 1996; Teleman et al. 1999; Lødrup 2002, inter alia). We review the evidence in the remainder of this section. Previous analyses differ in certain details, but there is general agreement on a number of basic syntactic and semantic facts. In essence, while examples like (5) are superficially consistent with an analysis as nothing more than two coordinated clauses, a monoclausal analysis is more plausible. Taken together, the evidence warrants a view of the SPC as a complex predicate in the sense of

1. Throughout this paper, examples taken from a corpus are cited using the following abbreviations: Gothenburg (Gothenburg Spoken Language Corpus), New novels (Bonnier’s romaner II corpus), Old novels (Äldre svenska romaner corpus), and Källtext (Källtext corpus). See §4.1 for a discussion of these corpora. Examples without citation are constructed.

A quantitative approach to the development of complex predicates 

Alsina et al. (1997), i.e., as a grammaticalized monoclausal construction with two verbal heads. To start out, we need to clarify that the verbs involved in the SPC do not show the features typically associated with auxiliary verbs. Crucially, SPC verbs like sitta carry finite morphology, are not phonologically reduced relative to their lexical counterparts, and do not have deficient paradigms. There is also no overt marker of embedding. Thus, an analysis of, for example, sitta “sit” as an auxiliary is ruled out. The standard reference grammar of Swedish (Teleman et al. 1999) characterizes the SPC as a construction in which certain semantically ‘light’ verbs, including sitta “sit”, ta “take”, and gå “go” are coordinated with a lexical verb. Another example is given in (6). (6) Mona satt och sydde i det blå rummet. Mona sat and sewed in the blue room “Mona was sewing in the blue room.”

(Teleman et al. 1999: 903)

As Teleman et al. point out, the light verb cannot receive primary stress without important changes in interpretation. If satt “sat” is stressed in (6) the sentence is likely to receive the interpretation that this is unexpected or in some way unusual. Stress on satt leads to the conceptual separation of two actions — here, sitting and sewing — which the construction portrays as integrated by default. Further evidence for monoclausality comes from the syntactic behavior of the SPC. First, unlike in main clause coordination, the second verb cannot have an overt subject. Example (7) is judged as questionable by Teleman et al. (1999). (7) ?Mona satt och hon sydde i det blå rummet. (Teleman et al. 1999: 903) Mona sat and she sewed in the blue room Intended meaning: “Mona sat and sewed in the blue room.” Secondly, the order of the two verbs in the SPC is fixed. They cannot be rearranged without disruption of the original meaning. (8) *Mona sydde och satt i det blå rummet. (Teleman et al. 1999: 903) Mona sewed and sat in the blue room Intended meaning: “Mona sewed and sat in the blue room.” A third syntactic indicator of monoclausality is that the two-verb sequence cannot be modified with the adverb både “both”. Doing so, as in (9), leads to the loss of the original, aspectual interpretation (seen in [6]). (9) *Mona både satt och sydde i det blå rummet. (Teleman et al. 1999: 903) Mona both sat and sewed in the blue room Intended meaning: “Mona both sat and sewed in the blue room.”

 Martin Hilpert & Christian Koops

Finally, syntactic movement phenomena suggest that pseudo-coordinated verbs exhibit greater syntactic integration than ordinary coordinated structures. For example, while the object of läsa “read” in (10) can occur outside of its default (post-verbal) position, this is not possible in (11). (10) Den där artikeln har jag suttit och läst hela dagen. that there article have I sat and read all day “That article I have been reading all day.” (Teleman et al. 1999: 903) (11) *Den där artikeln har jag skrattat och läst hela dagen. that there article have I laughed and read all day “That article he has laughed and read all day.” Like sitting and reading, laughing and reading can be co-temporaneous activities, but the SPC cannot be used to express the latter if the object is fronted. Teleman et al. (1999) note that the boundary between pseudo-coordination and regular coordination is fuzzy. Whether a given instance of, say, sitta in coordination with another verb is best analyzed as an instance of pseudo-coordination or as an instance of sitta used in canonical coordination cannot always be determined simply on the basis of positive evidence. However, as a rule of thumb, the more intervening elements occur between the two verbs, the weaker the conceptual union appears to be. Another way of stating the same generalization would be to say that the more the initial verb (e.g., sitta) is individually modified, the more it is understood to function as a lexical verb, whereas a lack of individual modification, as in (6) to (9), is suggestive of its light verb function. As we discuss in more detail below, this characterization is consonant with the changes seen in distributional behavior of the SPC in diachronic corpus data. It should be noted that while sitta is one of the most frequent SPC verbs, the phenomenon of pseudo-coordination goes far beyond what is discussed in this paper. The verbs that occur in the SPC can be thought of as a semi-closed class encompassing several subgroups which correspond to the crosslinguistically common sources for both light verbs and auxiliaries. Teleman et al. (1999) distinguish five broad classes. First are human posture verbs, especially sitta “sit”, stå “stand”, and ligga “lie”. The second group is formed by motion verbs like komma “come” and gå “go”. The verbs börja “start”, hålla på “hold on to, continue”, and sluta “stop” are subsumed under the category of “phase verbs” because they highlight a certain phase of an event. The fourth group consists of phrasal verbs with the copula vara “be” which encode polite requests. These are typically used in the imperative (e.g., var snäll “be nice” or var vänlig “be friendly”). The fifth category comprises verbs of communication, such as ringa “call” or skriva “write”.

A quantitative approach to the development of complex predicates 

3. Predictions and their corpus-linguistic operationalization This paper aims to test three predictions that follow from the hypothesis that repeated usage events over time shape grammar (Bybee et al. 1994; Bybee & Hopper 2001). This hypothesis contrasts with the view that language change occurs largely in the domain of language acquisition, where children instantaneously reanalyze the input they receive from adults (e.g., Lightfoot 1991). Our view of grammaticalization leads us to expect that linguistic structures, such as the argument structure of a verb which forms part of a grammaticalizing construction, undergo gradual changes. For example, hearing a verb in a novel context, e.g., with an unconventional argument frame, will not lead to an instantaneous and complete reanalysis. Rather, it will merely invite further usage of the verb in that way. As grammaticalization tends to correlate with frequency changes, the developmental course of a grammaticalizing element should be reflected in changing frequency patterns across historical periods. For the present analysis, these assumptions lead to the following predictions regarding the development of the SPC.

3.1 Prediction 1: Change in argument structure We assume that, as part of the grammaticalizing SPC, the verb sitta underwent a change in its argument structure. The reason for this lies in its changing semantics within the SPC, especially the gradual loss, or “bleaching out”, of the verb’s postural and locative meaning. We know from the synchronic data that the construction specifically brings out the aspectual meaning components of the verb, e.g., the sense of duration associated with sitting. At the same time, the postural and locative meaning components are backgrounded. Viewed over time, then, the more sitta functions as an aspectual marker, the less relevant its locative meaning becomes. This change in meaning and function should be reflected in the degree to which the verb is used together with prepositional phrases or other adverbials specifying the location or manner of sitting. In the following, we refer to any such modification of the verb’s lexical meaning as locative elaboration, and distinguish two broad types: cases where sitta is used with some form of locative elaboration and cases where it is not. Note that even in the lexical use of sitta locative elaboration is not obligatory. Both (12a) and (12b) are grammatical. (12) a. Jag har suttit vid skrivbordet nästan hela dan. I have sat at desk.the almost all day “I’ve sat at the desk almost all day.” b. Han satt en stund igen. he sat a while again “He sat (somewhere) for a while again.”

(New novels)

 Martin Hilpert & Christian Koops

Nevertheless, as part of the grammaticalizing SPC, the argument structure seen in (12b) should become more frequent. That is, we would expect cases like (12d) to increase in frequency relative to cases like (12c). (12) c. Jag har suttit vid skrivbordet och läst hela dagen. I have sat at desk.the and read all day “I’veˉsatˉatˉtheˉdeskˉandˉread all day.” d. Jag har suttit och läst hela dagen. I have sat and read all day “I have been reading all day.” In summary, to test the prediction of a change in the argument structure of sitta in the SPC we measure changes in the rate of locative elaboration (as defined above) in both the lexical and the SPC use of sitta. Non-elaborated usage should be more likely in the SPC overall and should increase with further grammaticalization of the SPC as a grammatical construction. On the other hand, when used outside of the SPC, as illustrated in (12a,b), the rate of locative elaboration should remain constant, because the original, lexical meaning of sitta does not change.

3.2 Prediction 2: Change in adverb placement Our second prediction rests on the synchronic observation that the SPC is syntactically more cohesive than verbs in canonical coordination. This leads us to expect a measurable increase in syntactic unity over time, as the construction developed from expressing canonical coordination to pseudo-coordination. An indicator of syntactic cohesion is the rate at which the coordinated constituents of the SPC are modified either individually or jointly by temporal adverbials like hela dagen “all day” where such modification occurs. There are three logical possibilities for the placement of adverbs like hela dagen in the SPC: before, between, or after the two verbs. When placed on the outside of the two-verb sequence, temporal adverbs are understood as modifying the event as a whole. On the other hand, adverbials placed between the two verbs — that is, following sitta — modify the first verb exclusively. We predict that the rate at which sitta is individually modified will decrease with increasing grammaticalization as a complex predicate because the view of the two events as integrated favors a form of modification which has both verbs in its scope. Thus, temporal adverbials should increasingly be placed outside the verbal complex. To illustrate, examples like (13a) should gradually give way to examples such as (13b). (13) a. Stock satt en stund tyst och tänkte över vad Marstrand hade sagt. Stock sat a while silent and thought over what Marstrand had said “Stock sat silent for a while and thought about what Marstrand had said.” (New novels)

A quantitative approach to the development of complex predicates 

b. Vi satt och pratade ett par timmar. we sat and talked a few hours “We sat and talked for a few hours.”

(New novels)

3.3 Prediction 3: Increase in the rate of object extraction The third prediction also derives from the SPC’s increase in syntactic unity. In those cases where the second verb occurs with a direct or prepositional object, that object should become available for extraction, which is not generally possible in coordinated structures (cf. §2). As the SPC increases in syntactic cohesion, extraction should become increasingly likely, and hence more frequent over time. To illustrate, examples like (14a) should become more frequent relative to examples like (14b). (14) a. Den där artikeln har jag suttit och läst hela dagen. that there article have I sat and read all day “That article I have been reading all day.” (Teleman et al. 1999) b. Jag har suttit och läst den där artikeln hela dagen. I have sat and read that there article all day “I have sat and read that article all day.”

4. Methodology 4.1 Corpus data We analysed four Swedish corpora, all of which are publicly accessible through the homepage of the Linguistics Department at the University of Gothenburg, Sweden (, date of access: 03/02/2006). The first corpus is the Källtext corpus, a collection of Old Swedish texts of about one million words. These texts comprise different genres, containing for example the religious treatise Saint Birgitta’s revelations, practical guidance literature such as Peder Månssons art of farming, and the five books of Moses. There is some uncertainty as to when some of the texts were composed exactly. It is, however, safe to say that most of them date to the period between 1300 and 1450. The second corpus, Äldre svenska romaner, consists of 57 novels from Swedish authors such as C.J.L. Almqvist, Victoria Benedictsson, and Hjalmar Söderberg. Taken together, they contain about 3.7 million words. The novels were originally published between 1839 and 1940. The third corpus, Bonnier’s romaner II, is a collection of Swedish novels, published in 1980 and 1981. It contains 4 million words. The fourth corpus, the Gothenburg Spoken Language Corpus, is a spoken corpus. It contains 1.4 million words of transcribed speech recorded in the 1990s.

 Martin Hilpert & Christian Koops

As will be clear from this description, the four corpora do not form a temporally continuous and internally homogeneous database. The represented time periods are not evenly spaced, and there are differences with regard to both genre and register. We cannot rule out the possibility that our results may be affected by these differences. The reason why we nonetheless base our analysis on these data is that, to our knowledge, no more suitable database is available at present. And rather than refraining from the use of these data altogether, we submit that the patterns we observe in the data have to be interpreted with the appropriate caution.

4.2 Corpus analysis Our quantitative methodology requires that the use of sitta be analyzed in its entirety, rather than on the basis of selected examples. Therefore, the first step of our analysis was an exhaustive extraction of all forms of sitta from each corpus. This was done through searches with several wildcards in order to accommodate the orthographical variety that is typically found in older corpus data. For example, for the paradigm of sitta our data include forms such as sato, sithiande, sithiä, and many others. All instances were retrieved, manually inspected, and entered into a database together with the sentence it occurred in. This procedure yielded four concordances of sitta, one for each corpus. Next, all instances of the verb sitta in each concordance were coded for several parameters. The first and most basic distinction was whether a given example potentially instantiates the SPC or not. Our operational criterion was whether sitta was used in some form of VP coordination. For example, cases analyzed as SPC exclude examples like (15a), but include presentative constructions with subject det “there”, as in (15b). (15) a. Snälla Blodwen, berätta var dom sitter och hur dom ser ut. dear Blodwen, tell where they sit and how they see out “Dear Blodwen, tell me where they sit and what they look like.” b. Det satt en liten fågel vid min fot och kvittrade så vackert. there sat a little bird at my foot and chirped so beautifully “A little bird sat at my foot and chirped so beautifully.” Second, all instances of sitta were coded for the presence of modifiers falling into the category of locative elaboration (cf. §3.1). We counted as forms of locative elaboration any modification of sitta by means of an adverb or prepositional phrase specifying the location or manner of sitting. For instance, the prepositional phrase vid min fot “at my foot” in example (15b) elaborates the locative sense of the sitta. Third, we took account of all instances in which SPC sitta is modified by temporal adverbials. Besides temporal adverbials such as hela kvällen “all evening long” we also included a small number of other adverbs, for example adverbs indicating psychological

A quantitative approach to the development of complex predicates 

states such as fundersam “pensive” which are not clearly spatial in nature. We coded the placement of these adverbials in terms of two mutually exclusive environments: external to or between the two verbs, as in (16a) and (16b), respectively. (16) a. Hon har suttit och sett på dig hela kvällen. she has sat and looked at you all evening “She’s been looking at you all evening.”

(Old novels)

b. Magistern satt fundersam ochtittade på vägen. magister sat pensive and looked at path.the “The teacher pensively looked along the path.”

(Old novels)

Finally, we coded all instances of sitta in SPC contexts with regard to whether the second verb can take a direct or prepositional object, and if so, whether the object of the verb is in-situ or not. Example (17a) shows an extracted prepositional object, while the object of (17b) is in-situ. (17) a. Me vet ni vad jag sitter och tänker på? but know you what I sit and think about “But do you know what I am thinking about?” b. Jag satt och tänkte på Brita. I sat and thought about Brita “I was thinking of Brita.”

(Old novels)

(Newˉnovels)

5. Results and discussion Table 1 presents the absolute and relative frequencies of sitta inside and outside of potential SPC contexts for each of the four periods. The frequencies show that the use of sitta in these contexts has increased over time, from 18.4% of all instances in the Källtext corpus to about 40% in the later corpora. This frequency increase in itself suggests an underlying qualitative change. Table 1. Token figures and relative frequencies of the SPC in four corpora of Swedish. Källtext

Old novels

1M Size SPC

New novels

3.7 M

Gothenburg

4M

1.4 M

n

%

n

%

n

%

n

%

96

18.4

2,053

38.6

2,138

41.5

278

40.1

non-SPC

426

81.5

3,269

61.4

3,012

58.5

416

59.9

Totals

522

100

5,322

100

5,150

100

694

100

 Martin Hilpert & Christian Koops

A closer inspection of the attested use of sitta in the Källtext data suggests that an SPC-like construction already existed in the 14th century. An example is given in (18). (18) Ther sato nokre kompana oc drukko oc lifdho i ofwerflødhlikheth. there sat some friends and drank and lived in abundance “There sat some friends and drank and lived in abundance.” (Källtext) Given the problems with identifying pseudo-coordination strictly on the basis of positive evidence, one might attempt to explain cases like (18) simply as canonical coordination. Indeed, all of the examples in the Källtext corpus express actions that are either conventionally carried out in a sitting posture, such as eating or drinking, or are at least fully compatible with a sitting posture, such as talking. However, there is qualitative evidence which suggests that the construction had already acquired a grammatical function at this stage. Consider (19), which describes two temporally intersecting events. (19) Tha han kom til qwinnan oc sat och taladhe med henne, tha when he came to woman and sat and talked with her then frestade diæfwllin honom … tempted devil him

(Källtext)

“When he came to the woman and sat and talked to her, the devil tempted him…” In this example, the first event (talking) provides the temporal background for the second event (the devil’s temptation). This use of sitta in (19) is characteristic of the discourse function of progressive and other imperfective aspect constructions (Comrie 1976: 33). While the backgrounded event is given an imperfective construal by means of sitta, the foregrounded event takes perfective marking. Examples like (19), then, suggest that the grammaticalization of the SPC was already underway at the earliest period. Besides these qualitative observations, there is also quantitative evidence showing that the SPC already existed as a grammaticalized construction at this state. Figure 1 shows the overall frequency of locative elaboration of sitta inside or outside of SPC contexts (for the absolute frequencies, please see the Appendix). There is a statistically significant preference for zero-elaboration when sitta occurs in an SPC context (df=1, χ2=20.73, p<.001, absolute frequencies are given in the Appendix). In other words, there is a measurable preference for sitta to be used in its “bare” form when in coordination with another verb. This asymmetry in the argument structure can only be captured by saying that sitta was used in its unelaborated form specifically when used together with another verb. And this is tantamount to saying that a sitta-based grammatical construction existed. If sitta already formed part of a complex predicate at this early stage, the next question is whether there is evidence of further grammaticalization. Any additional development in the same direction should be reflected in an increase in the asymmetry

A quantitative approach to the development of complex predicates 

Relative frequency

1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0

unelaborated elaborated

SPC

Non-SPC

Figure 1. SPC/non-SPC and elaboration/non-elaboration in the Källtext corpus. 0,8

Relative frequency

0,7 0,6

sitta ochV

sitta Ø

0,5 0,4 0,3 0,2 0,1 0,0

Källtext

Old novels

New novels

Gothenburg

Figure 2. Relative frequency of zero-elaboration of sitta inside and outside of the SPC.

in Figure 1. The difference between the ratio of non-elaborated and elaborated instances of sitta in the SPC and the same ratio outside of the SPC should become more dramatic. Figure 2 presents these relative frequencies (for the absolute frequencies, please see the Appendix). The shaded bars show that when sitta is used as a lexical verb it occurs only about 10% of the time without some form of locative elaboration. This rate is relatively constant across the four corpora.2 Indeed, we would not expect any changes here because the argument structure of lexical sitta and remained unchanged. By contrast, the white

2. This suggests that the four corpora are, indeed, comparable in this respect.

 Martin Hilpert & Christian Koops

bars indicate a steady increase in “bare” uses of sitta in the SPC. These changing preferences suggest that there was a change in argument structure of sitta that is specific to its function within the SPC. As predicted, in the course of continuing grammaticalization, there is an increasing tendency to use sitta in its most compact form, which backgrounds its locative meaning by not further specifying the location or manner of sitting. The differences between the four corpora are statistically significant overall (df=3, χ2=71.66, p<.001). In keeping with our earlier observation that the data need to be interpreted cautiously, we need to point out that time is only one parameter by which the four corpora differ. Still, the fact that significant differences exist is suggestive of a gradual change. We now turn to our second prediction, viz. that adverbials should increasingly be used to modify the entire verbal complex and therefore occur before or after the two verbs, rather than between them. As not all instances of the SPC contain adverbial modification, this parameter can only be applied to those examples which actually are adverbially modified (about 15% of the data). Again, the hypothesis is borne out. Figure 3 shows that the relative frequency of adverbials placed between the two verbs and adverbials external to the verbal complex changed in the expected direction (for the absolute frequencies, please see the Appendix). The asymmetry in the distribution of intervening and external adverbials is statistically significant across the four corpora, (df=3, χ2=18.35, p<.001).

Relative frequency

0,4

intervening adverbials external adverbials

0,3

0,2

0,1

0,0

Källtext

Old Novels

New Novels

Gothenburg

Figure 3. Internal and external adverbials in the SPC.

Our third prediction was that an increase in the syntactic unity of the SPC should be reflected in a greater rate of extracted objects from the second, lexical

A quantitative approach to the development of complex predicates 

verb. Again, only a subset of the data is relevant, as not all verbs take objects. About 40% of all retrieved examples contain a verb that takes either a direct or prepositional object. In cases where the second verb is transitive, the object can be extracted for the purpose of topicalization, or in order to function as the head of a relative clause. Over time, we expect speakers to make more use of this possibility. This prediction is borne out, as shown by Figure 4 (as previously, all absolute numbers are listed in the Appendix). 1 0,9

Relative frequency

0,8

extracted

in-situ

0,7 0,6 0,5 0,4 0,3 0,2 0,1 0

Källtext

Old Novels

New Novels

Gothenburg

Figure 4. Increase of extracted objects of V2 in the SPC.

There is a significant difference in the relative frequency of extraction phenomena across the four corpora (df=3, χ2=33.1, p<.001). As we expected, extraction is very uncommon in the older data. In the oldest period, thirty-one examples occur with object-taking verbs, but none of these are extracted. However, extractions are attested at the second stage, as shown in (20). (20) Det var kanske den bibeln husmodern satt och läste i, då it was maybe that bible house.mother.the sat and read in when de kom … they came “Maybe it was that bible that mother sat and read when they came … ” It needs to be pointed out that even in synchronic data the relative frequency of extracted objects is still not very high overall. Only in about 7% of the possible cases does extraction occur.

 Martin Hilpert & Christian Koops

6. Conclusion The observed tendencies allow the following conclusions. The Swedish Pseudocoordination construction with sitta has, even after it established itself as a grammatical construction, undergone a number of gradual changes. As part of the SPC, the verb sitta has changed its argument structure, and the construction showed in increase in syntactic cohesion. The present paper also bears on a disputed point brought up in the study of light verbs. Butt & Lahiri (1998) compare complex predicates from Sanskrit, Old Bengali, Old Hindi, and Vedic to corresponding constructions in present-day Indo-Aryan languages. They find that their structure has remained essentially the same in spite of three millennia of language change. Butt & Lahiri conclude that auxiliaries are more prone to change than light verbs. While the process of grammaticalization turns lexical verbs into auxiliaries through semantic bleaching and morphosyntactic reduction, which eventually leads to affixation and complete deletion, light verbs are claimed to be historically stable (1998: 51). Butt & Lahiri explicitly reject proposals by Hook (1991, 1993), who proposes that light verbs underlie the same principles as other grammaticalizing elements. The present analysis covers a far shorter period of time than the study by Butt & Lahiri, but yields tentative counterevidence against their claim. It has been shown that sitta displays characteristics of a light verb, as it takes on a type of aspectual meaning and clearly does not undergo auxiliation. Contra Butt & Lahiri, the data suggest that sitta has not been historically stable, but was instead subject to gradual change. So far this change has only affected the argument structure of sitta as part of the SPC; we do not know whether it will ultimately lead to morphophonemic reduction. However, the development so far does not suggest that light verbs behave any differently than auxiliaries, a conclusion also reached by Hook & Pardeshi (2006). As a final conclusion, we find that the traditional approach of studying grammaticalization through crosslinguistic comparisons and grammaticality judgments can be usefully supplemented by an approach based on quantitative data whenever this is possible. While many languages have no written record of great time depth, the languages for which such data are available provide opportunities to test hypotheses that not only bear on language-specific questions but also on general principles of language change. Ultimately, the richest accounts of grammatical phenomena will be those that are based on different types of evidence.

A quantitative approach to the development of complex predicates 

Appendix Table 2. Elaborated and non-elaborated instances of sitta in the SPC and elsewhere. Lexical sitta

spc Unelaborated Källtext

Elaborated

Unelaborated

Elaborated

30

66

53

373

Old Novels

879

1,174

440

2,829

New Novels Gothenburg

919 198

1,219 80

398 36

2,614 380

Table 3. Intervening and external adverbials in the SPC. Intervening adverbials Källtext

External adverbials

All SPC examples

9

0

96

Old Novels

243

31

2,053

New Novels Gothenburg

216 34

37 17

2,138 278

Table 4. Extracted and in-situ objects of V2 in the SPC. Extracted

in-situ

Källtext

0

31

Old Novels

7

987

New Novels Gothenburg

12 8

996 105

References Aikhenvald, A. 2006. Serial verb constructions in a typological perspective. Serial Verb Constructions: A Cross-linguistic Typology, A. Aikhenvald & R.M.W. Dixon (Eds), 1–87. Oxford: OUP. Alsina, A., Bresnan, J. & Sells, P. (Eds). 1997. Complex Predicates. Stanford CA: CSLI. Alsina, Alex. 1997. A Theory of Complex Predicates: Evidence from causatives in Bantu and Romance. Complex Predicates, ed. by Alex Alsina, Joan Bresnan, & Peter Sells, 203–246. Stanford: CSLI. Austin, Peter. 1998. Crow is sitting chasing them: Grammaticalisation and the verb to sit in the Mantharta languages, Western Australia. In Case, Typology and Grammar: In honour of Barry J. Blake, A. Siewierska & J.J. Song, 19–36. Amsterdam: John Benjamins.

 Martin Hilpert & Christian Koops Bowern, C. 2008. The diachrony of complex predication: An introduction. Diachronica 25(2), 161–85. Butt, M. 1995. The Structure of Complex Predicates in Urdu. Stanford CA: CSLI. Butt, M. & Lahiri, A. 1998. Historical stability vs. historical change. Ms. Konstanz: University of Konstanz. Bybee, J.L. 1985. Morphology: A Study of the Relation between Meaning and Form. Amsterdam: John Benjamins. Bybee, J.L., Perkins, R.D. & Pagliuca, W. 1994. The Evolution of Grammar: Tense, Aspect and Mood in the Languages of the World. Chicago IL: University of Chicago Press. Bybee, Joan L. & Paul Hopper. (Eds). 2001. Frequency and the Emergence of Linguistic Structure. Amsterdam: John Benjamins. Comrie, B. 1976. Aspect. Cambridge: CUP. Di Sciullo, A.M. & Thomas Rosen, S. 1990. Light and semi-light verb constructions. In Grammatical Relations: A Cross-theoretical Perspective, K. Dziwirek, P. Farrell & E. Mejías-Bikandi, 109–125. Stanford CA: CSLI. Foley, W. & Olson, M. 1985. Clausehood and verb serialization. In Grammar inside and outside the clause, J. Nichols & A. Woodbury (Eds), 17–59. Cambridge: CUP. Gahl, S. & Garnsey, S. 2004. Knowledge of grammar, knowledge of usage: Syntactic probabilities affect pronunciation variation. Language 80(4): 748–775. Givón, T. (in this volume). Multiple routes to claus union! The diachrony of verb phrases. Harris, A.C. 2008. Light verbs as classifiers in Udi. Diachronica 25(2), 213–41. Heine, B., Claudi, U. & Hünnemeyer, F. 1991. Grammaticalization: A Conceptual Framework. Chicago IL: University of Chicago Press. Heine, B. & Kuteva, T. 2002. World Lexicon of Grammaticalization. Cambridge: CUP. Hook, P.E. 1991. The emergence of perfective aspect in Indo-Aryan Languages. In Approaches to Grammaticalization, B. Heine & E.C. Traugott, 59–89. Amsterdam: John Benjamins. Hook, P.E. 1993. Aspectogenesis and the compound verb in Indo-Aryan. Complex Predicates in South Asian Languages, M. Verma (Ed.), 97–113. Delhi: Manohar. Hook, P. & Pardeshi, P. 2006. Are vector verbs eternal? Paper presented at the 11th Biennial Rice Linguistics Symposium: Intertheoretical Approaches to Complex Verb Constructions. Hooper, J.B. 1976. Word frequency in lexical diffusion and the source of morphophonological change. In Current Progress in Historical Linguistics, W. Christie (Ed.), 96–105. Amsterdam: North Holland. Hopper, P.J. & Traugott, E.C. 2003. Grammaticalization, 2nd Edn. Cambridge: CUP. Kuteva, T. 2001. Auxiliation - An Enquiry into the Nature of Grammaticalization. Oxford: OUP. Lightfoot, D.W. 1991. How to set Parameters: Evidence from Language Change. Cambridge MA: The MIT Press. Lødrup, Helge. 2002. The syntactic structures of Norwegian pseudocoordinations. Studia Linguistica 56: 2.121–143. Svorou, S. 1994. The Grammar of Space. Amsterdam: John Benjamins. Teleman, U., Hellberg, S. & Andersson, E. 1999. Svenska Akademiens grammatik. Stockholm: Norstedts. Wiklund, A.-L. 1996. Pseudocoordination is subordination. Working Papers in Scandinavian Syntax 56: 29–54.

Elements of complex structures, where recursion isn’t The case of relativization* Masayoshi Shibatani Rice University

In their recent work, Hauser, Chomsky and Fitch (2002:1569) suggest that recursion “is the only uniquely human component of the faculty of language”. In both generative and typological studies, the relativization site has been considered to be one of the places where recursion of sentences takes place. This paper examines a number of wide-spread patterns of relativization around the globe and argues that what have been identified as relative clauses/sentences are in fact nominalized entities, lacking some crucial properties of both full clauses and sentences. It is furthermore shown that these nominalized forms are neither syntactically nor semantically subordinate to, or dependent on, the nominal head they modify.

*Work reported here was supported in part by the National Science Foundation grant BCS-0617198. I am grateful to the Institute for Advanced Study and the Research Centre for Linguistic Typology at La Trobe University in Melbourne, Australia for the hospitality extended to me while I was revising this paper. Special thanks are due to Sasha Aikhenvald, Noritaka Fukushima, Albert Alvarez, Hiroshi Hayashi, and Roberto Zariquiey for the helpful discussions on the topics dealt with in this paper and for providing me with relevant examples from their extensive knowledge of the languages identified below. The following list includes other colleagues who provided data on the languages of their expertise, for which I am very grateful. Examples not identified for their sources are either due to the courtesy of the following individuals or based on my personal knowledge: Alexandra Aikhenvald (Russian), Michael Colley (Brazilian Portuguese), Jaime Daza (Bolivian Quechua), Bjarke Frellesvig (Danish), Telma Kaan (K’ichee’), Noritaka Fukushima (Spanish), Albert Alvarez (Yaqui, Spanish, French), Hiroshi Hayashi (French), Martin Hilpert (Swedish), Chenglong Huang (Qiang), František Kratochvíl (Czech),Yu Kuribayashi (Turkish), Thomas Larsen (K’ichee’), Naonori Nagaya (Tagalog), Akio Ogawa (German), Stella Telles (Brazilian Portuguese), Vera Podlesskaya (Russian), Roberto Zariquiey (Spanish, Quechua).

 Masayoshi Shibatani

1. Chomskyan notation for recursion and syntactic structures As far as I can see, there are two negative consequences of the traditional Chomskyan notation for recursion of the following form or its variants accounting for complex syntactic structures. (1)

a. S → NP VP b. NP → (D)N′ c. N′ → N(S′) d. VP → V (S′) e. S′ → (C) S

One is the implication that what is embedded within a NP or a VP is the same object as the main clause, namely a sentence. The other, related assumption is that a full sentence with all its arguments underlies a clausal complement and a modification structure such as a relative clause. These have had a profound effect on our thinking about the nature of complex syntactic structures and their analysis in both synchronic and diachronic dimensions. Synchronically, we have been taught that relative clauses and verb complement constructions, for example, have the following underlying structures:

(2) a. Relative clause NP D

N′ N

the

dogi

S′ C

S

that

John saw the dogi/iti

(the dog that [John saw])

b. Verb complement VP V

S′ C

try

S I read his book

(I try [to read his book])

Elements of complex structures, where recursion isn’t 

Many interesting research results have been achieved based on the assumptions embodied in the Chomskyan recursive rules and the syntactic structures they countenance. For example, Keenan and Comrie’s (1977) seminal work on the universal constraints on relativization is predicated on the assumption that underlying relative clause structures have the form of (2a) above. Such an assumption allows them to talk about “the NP in the restricting sentence that is coreferential with the head noun as the NP relativized on” (64; emphasis added), and to establish the concept of accessibility to relativization formulated in terms of the grammatical relations of the NPs relativized on. The achievements made in this work, however, crucially depend on the assumption noted above. Once this assumption and other attendant assumptions are abandoned, the achievements would turn out to be no more than a sandcastle. We will see that this is the case below. The other, diachronic arena is also fraught with ideas that a sentence might be integrated with another one giving rise to such complex structures as relative clauses, clausal complements, subordinate adjuncts, and serial verb constructions. For example, Hopper and Traugott (1993:169) note that “[f]rom the point of view of language change, the initial formation of a complex structure involves the combining into one integrated structure of two separate and autonomous nuclei [sentences] that are mutually relevant” as depicted in (3) and (4) below:

(3) S1 <=> S2 (4)

S1

S S2

Whereas Hopper and Traugott (1993) are a bit more careful in their description of the transition from the paratactic to other more integrating patterns of clause combination along the cline of parataxis > hypotaxis > subordination, others have been less so. For example, Heine and Kuteva (2007:214) have recently suggested two channels through which clause subordination arises, namely “[e]ither Expansion, that is, the reinterpretation of a nominal as a clausal (propositional) participant, or via the integration of two independent sentences within one sentence” (emphasis added), and have sketched the two patterns of development in the following manner:

(5) a. S [NP] > S1 [S2] Expansion (complements and adverbial clauses)

b. [S1 + S2] > S1 [S2] Integration (relative clauses) While it is true that a sentence may consist of two or more sentences, as at the paratactic stage shown in (4), where the two subparts are not structurally integrated, the transition from a paratactic structure to a more integrating hypotactic and subordinate one seems to require greater cognitive processing, not shown in (5), than the formal

 Masayoshi Shibatani

hierarchical integration of two sentences into a complex structure as shown in (5). This is easy to see if we look at what is involved in converting a direct quote into an indirect one in a language, such as (Old) Japanese, which has numerous predication features (e.g., discourse particles, honorifics, evidential- and tense-marking) distinguishing between a sentence and a(n embedded) clause. Indeed, the essence of clause integration seems to be the cognitive ability to convert a sentence into a non-sentential nominal object, which can then be legitimately integrated into a matrix sentence. In what follows, I would like to show this largely on the basis of relative clause formation in Austronesian languages and others, where a full clause, let alone a sentence, is not involved, contrary to what is suggested by the syntactic structure shown in (2a) or the schematic representation such as (5b) (see Deutscher in this volume for a very similar view). The discussion below is basically synchronic but the plausibility of diachronic speculations based on comparative data from diverse languages such as Heine and Kuteva (2007) and others rests on the correct understandings of the synchronic structures in question.

2. Austronesian relativization Despite the reduction in the morphological contrast and even a total loss of such contrast in some dialects, Sasak of Lombok Island in eastern Indonesia is typical of Western Austronesian languages in maintaining the structural contrast between so-called Actorfocus (af) and Patient-focus (PF) constructions.1 This is clear from the way relativization works in these dialects. That is, consistent with other relevant Austronesian languages, only the primary argument, referred to variably in the literature as “topic”, “subject”, “pivot” or “trigger”, can be relativized on — the fact that has been construed as the “subjects-only” constraint by Keenan and Comrie (1977) and that underlies one of the proposed universal constraints on relativization formulated as: “subjects are universally the most relativizable of NPs” (Keenan 1985:158) or its weaker version: “all languages can relativize Subjects.” (Comrie & Keenan 1979:652) (6) Pancor Ngeno-Ngené Sasak a. dengan mame ino mantok loq Ali (af) person male that N.hit art Ali ‘That man hit Ali.’ a′. dengan mame [si Ø mantok loq Ali] batur=meq person male rel N.hit art Ali friend=2sg ‘That man who hit Ali is your friend.’ (Topic A relativized) . In this paper the term“Western Austronesian languages” is used as a cover term embracing both Western Malayo-Polynesian languages (of the Philippines, Indonesia, Malaysia, and Madagascar) and Formosan languages (of Taiwan). See Shibatani (2008) for the details of Sasak relativization.

Elements of complex structures, where recursion isn’t 

a′′. *Loq Ali [si dengan mame ino mantok Ø] batur=meq art Ali rel person male that N.hit friend=2sg ‘Ali, whom that man hit, is your friend.’ (Non-Topic P relativized) b. Loq Ali pantok=na siq dengan mame ino (pf) art Ali Ø.hit=3sg by person male ‘That man hit Ali.’ b′. loq Ali [si Ø pantok=na siq dengan mame ino] batur=meq art Ali rel hit=3sg by person male that friend=2sg ‘Ali, whom that man hit, is your friend.’ (Topic P relativized) b′′. *dengan mame [si Ali pantok=na Ø] batur=meq person male rel Ali hit=3sg friend=2sg ‘The man who hit Ali is your friend.’ (Non-Topic A relativized) The gaps in the relative clauses above indicate the positions of the relativized NPs in the Keenan-Comrie approach. Comrie and Horie (1995) and Comrie (1998) argue, on the basis of the apparent parallelism between relative clauses with gaps and ordinary sentences with similar gaps in Japanese seen below, that relative clauses (in Japanese) are no different from ordinary sentences with anaphoric gaps, and that rc gaps are not the ones created by extraction or movement of the relativized NPs as in the standard generative analysis. (7) Japanese a. kore=ga [Ø kinoo Ø katta] hon desu. this=nom yesterday bought book cop ‘This is the book that (I) bought Ø yesterday.’ b. Ø kinoo Ø katta. yesterday bought ‘(I) bought (it) yesterday.’ While (7b) is a perfect independent sentence of Japanese that answers a question (such as “Have you bought the book already?”), the parallelism between relative clauses and independent sentences seen above is deceptive, and a similar situation may not obtain in other languages. For example, to the question in (8a) below, the appropriate answer in Pancor Ngeno-Ngené Sasak would be either (8b) with full pronouns or (8c) with pronominal clitics, while the Japanese answer would have gaps for “I” and “it”. (8) Pancor Ngeno=Ngené Sasak a. Kumbeq=meq buku=no? what.do=2 book=that ‘What did you do with that book?’

 Masayoshi Shibatani

b. Aku nulak-ang ia tipak perpustakaan I return-appl it to library ‘I returned it to the library.’ c. Ku=nulak-ang-e tipak perpustakaan 1=return-appl-3 to library ‘I returned it to the library.’ The corresponding relative clause, however, cannot contain the object clitic, indicating that the relative gap here is an obligatory gap. (9)

Pancor Ngeno-Ngené Sasak Buku [si ku=tulak-ang=*e/Ø tipak perpustakaan]=no bagus book rel 1=return-appl=3 to library=that interesting ‘The book that I returned to the library was interesting.’

Even in Japanese, the two gaps seen in the relative clause in (7a) differ in that while the first gap corresponding to the subject nominal can be overtly expressed, the second one corresponding to the object nominal coreferential with the head cannot in any form. This shows that the parallelism Comrie draws between independent sentences and relative clauses does not in fact obtain and that rc gaps are different from those anaphoric gaps created under discourse conditions. Just as identifying relative clauses as ordinary sentences is mistaken, labeling markers such as si in Pancor Ngeno-Ngené Sasak and its equivalents in many other languages as relativizers or relative clause markers, as we have done above, is also misleading. Expressions headed by si and its equivalents (siq, saq, siq-saq) in the Sasak dialects occur in a wide range of modification functions shown below, ranging from both nominal and verbal complements and subordinate adverbial expressions, which do not necessarily have a gap like the one found in a relative clause. (10) Pancor Ngeno-Ngené Sasak a. buku [si Ø ne=bace isiq loq Ali] book nmz 3=Ø.read by art Ali ‘the book that Ali read’ b. suara [si ne=ngerontok lawang loq Ali] sound nmz 3=knock door art Ali ‘sound of Ali knocking on the door’ c. berita [si angku=n loq Ali beruq merariq] news nmz way=3 art Ali recently marry ‘the news that Ali recently got married’ d. Aku lupaq [si angku-n loq Ali wah mbilin kota=no] I forget nmz way=3 art Ali perf leave town=that ‘I forgot that Ali had left the town.’

Elements of complex structures, where recursion isn’t 

e. waktu [si ku=masih sekolah]=no … time nmz 1=still school=that ‘At the time when I was still going to school…/When I was going to school…’ f.

Ali te-semateq [si=ne lekaq léq rurung] Ali pass-kill nmz=3 walk loc street ‘Ali was killed when/while he was walking in the street.’

A more appropriate label for si in Pancor Ngeno-Ngené Sasak is “nominalization marker” glossed as nmz in the examples above. Then our understanding of Sasak relativization takes a different turn, namely relativization in Sasak — and many other languages, as we shall see below — represents one of the modification functions of nominalized expressions rather than a phenomenon involving full clauses or full sentences subordinated to the head nominal, as in the traditional generative and typological analysis. Indeed, si and its equivalents in other Austronesian languages (such as yang in Bahasa Indonesia) mark nominalizations of even such items as demonstratives and ordinal numbers in the same manner as it marks nominalized clauses, as shown by the nominalization marker saq in Puyung Meno-Mené Sasak below: (11) Puyung Meno-Mené Sasak a. [saq ino] baru nmz that new ‘That one is new.’ b. [saq pertame] mame kance [saq kedue] nmz first male and nmz second

nine (speaking about one’s children) female

‘The first one is male and the second one is female.’

c. Gitaq [saq Ø nyenke=n tokol leq bucu=no] look nmz prog=3sg sit loc corner=the ‘Look at the one sitting in the corner.’ d. [saq Ø nyenke=n tokol leq bucu=no] amaq=k nmz prog=3sg sit loc corner=the father=1sg ‘The one sitting in the corner is my father.’ It is my father who is sitting in the corner.’ e. Amaq=k [saq Ø nyenke=n tokol leq bucu=no] father=1 nmz prog=3sg sit loc corner=the ‘My father is the one sitting in the corner.

 Masayoshi Shibatani

f.

Kamu amaq=k you father=1sg ‘You are my father.’

The above examples also present clear evidence that si, saq and others indeed head nominalized expressions functioning as arguments or nominal predicates in the same manner as simple nouns in (f). To summarize the relative clause in Western Austronesian languages is an appositive construction consisting of a nominal (corresponding to a head nominal) and a nominalized expression with a gap (corresponding to a relative clause) juxtaposed as in (12) below. As an appositive construction, the head nominal identifies the entity referred to by the nominalized expression functioning as a relative clause (see further discussion on this point in section 8 below). The nominalized expression, in turn, functions as a modifier of the head nominal. (12)

Pancor Ngeno-Ngené Sasak buku [si beng=ku iye Ø]=no book nmz give=1 he =that ‘the book that I gave him’

Modification of a noun by another nominal element is not at all rare, as seen in noun compounds (e.g., gold watch, songbird) and in the use of participle forms as modifiers (e.g., singing bird, a bird singing in the bush).

3. Austronesian nominalizations The reason that I prefer using the term “nominalization marker” above to the more conventional “nominalizer” for Sasak particles such as si and saq heading nominalized forms is that Austronesian nominalizations are in general formed without any such marker. For example, Formosan language Mayrinax Atayal nominalizes (part of) a sentence without any additional marker. (13) Mayrinax Atayal (based on Huang 2002) a. yakaat m·inÒuwah cuʔhisaʔ kuʔ naßakis neg AF�perf�come yesterday nom.ref old.man ‘The old man didn’t come yesterday.’ b. kiaʔ ʔiʔ m-aniq kuʔ [yakaat m·inÒuwah cuʔhisaʔ] prog lin af-eat nom.ref neg af�perf�come yesterday ‘The one who didn’t come yesterday is eating (there).’

Elements of complex structures, where recursion isn’t 

c. kiaʔ ʔiʔ m-aniq kuʔ cuqliq ka′ [yakaat m�in�uwah cuʔhisaʔ] prog lin af-eat nom.ref person lin neg af�perf�come yesterday ‘The person who didn’t come yesterday is eating (there).’ As the above examples show, in Mayrinax Atayal nominalized form without any nominalization marker functions both as an argument marked by the nominative particle (13b) and as a nominal modifier (or rc) linked to the head nominal (13c). The parallel pattern obtains in Tagalog, as shown below: (14) Tagalog a. Hindi d·umÒating ang matanda-ng lalaki kahapon neg come�af� top old-lin man yesterday ‘The old man didn’t come yesterday.’ b. K�um�a-kain doon ang [hindi d·umÒating kahapon] dup�AF�-eat there top neg come�AF� yesterday ‘The one who didn’t come yesterday is eating there.’ c. K�um�a-kain doon ang tao-ng [hindi d·umÒating kahapon] dup�AF�-eat there top person-lin neg come�AF� yesterday ‘The person who didn’t come yesterday is eating there.’ The nominalization markers in Sasak dialects, which seem to be a later development, mark what has been nominalized as such, much like Chinese de and Japanese no. Such nominalization markers may not occur in certain contexts, as in the case of Chinese de and Japanese no, or may be optional like Sasak si, siq, etc. What is responsible for argument nominalization in Western Malayo-Polynesian and Formosan languages is the so-called Austronesian focus morphology, which profiles the grammatical role of the argument nominalized. While the same morphology is seen in the predicate of a sentence (cf. (14a) and (14b) above), it is likely that the original function of Austronesian focus morphology lied in argument nominalization (see below on the development of focus-marked sentence predicates). Thus, the so-called Actor focus (af) affix derives an Actor nominalization, which typically refers to an agent that performs some action. Similarly, the pf affixe derives a nominal expression typically referring to what is affected. In the same vein, Locative (lf) and Instrumental (If) affixes derive nominals designating a place where a certain action takes place and an instrument used to perform the named action, respectively. (15) Mayrinax Atayal (based on Huang 2002) a. m-aquwas ku irawaiŋ=mu (Af construction) AF-sing nom.ref friend=1sg.gen ‘My friend is singing.’

 Masayoshi Shibatani

a′. ßaq-un=mu kuʔ [m-aquwas] kaʔ know-pf=1sg.gen nom.ref AF-sing lin

hacaʔ that

(Actorˉnominalization)

‘I know that singer/one who is singing there.’

b. ma-hnuq kuʔ [ß-in-ainay nukʔ naßakis] AF-cheap nom.ref buy�pf.realis�buy gen.ref old.man ‘What the old man bought was cheap.’

(Patient)

c. ɣaɣhapuyan kuʔ [naniq-an cuʔ ßuŋaʔ nkuʔ kitchen nom-ref eat-lf acc.nonref yam gen.ref

ʔulaqiʔ] child

(Locative)

‘The kitchen is (the place) where the child eats yam.’

d. kaa ptiq-ani kuʔ [pa-patiq=mu] neg.imp write-if.imp nom.ref if-write=1sg.gen ‘Don’t write with my pen/thing to write with.’

(Instrumental)

In Sasak and other languages in which focus morphology is reduced, there can be only two or three types of argument nominalizations, as we saw in Sasak above. In any event, it is clear that the gap contained in a noun modifier (or rc) in Austronesian languages is the one that has been created in the process of argument nominalization rather than in the relativization process, which, as noted above, simply juxtaposes a (head) nominal and an argument nominalized expression with a gap in appositive syntagm.2

. Our position regarding the relative clause constructions in Austronesian differs from that of Foley’s (1976), who analyzes RC and participial constructions as instances of the [Adjunct + N] construction, which is defined as “non-nominal modifier of a head noun within a noun phrase” (13; emphasis added) distinguishing them from the [Noun + Noun] construction type, which involves “nominalizations, gerunds and possessive phrases” (69). LaPolla with Huang (2003:225) seems to take a similar position (see also Huang 2008). Though the exact types of construction that fall under their [NP + N(P)] are unclear (nominalized RCs in Qiang are included in this type), they seem to exclude finite relative clauses from this construction type. Our position is that all these modification structures, including finite relative clauses, are basically of the [Noun + Noun] type.

Elements of complex structures, where recursion isn’t 

5. Argument nominalizations in other languages The pattern of argument nominalization and the role of the focus morphology in Western Austronesian languages seen above actually are not entirely unique to this language group. Indeed, a fairly large number of languages around the world have argument nominalizations of similar type. Many Tibeto-Burman languages have morphology distinguishing between agent nominalization, patient nominalization, and instrument nominalization, as shown in (a) forms below:3 (16) Northern Qiang (Ronghong) (Huang 2008:194–196) Agent nominalization a. fa ɕupu guә-m (equivalent to Austronesian AF) clothes red wear-nmz ‘one wearing red clothes’ b. [fa ɕupu guә-m] tɕymi the: clothes red wear-nmz child that.cl ‘that child who wears red clothes’ (17) Patient nominalization a. [qa (-wu) khe]-tɕ 1sg-agt cut-gen ‘one I am cutting’

(equivalent to Austronesian pf)

b. [qa (-wu) khe]-tɕ sәf tho-zgu 1sg-ag cut-gen tree that-cl ‘the tree that I am cutting’ (18) Instrumental nominalization a. pies khukhu-s meat slice-nmz ‘what (is used) to slice meat.’

(Appositive rc)

(Appositive rc)

(equivalent to Austronesian if)

b. tse: [pies khukhu-s] xtşepi ŋuә that:cl meat slice-nmz knife cop ‘That is the knife that is used to slice meat.’

(Appositive rc)

The Rhonghon dialect of Qiang marks agent nominalization by the suffix -m and instrumental nominalization by -s. Patient nominalization, on the other hand, involves

. Whether these processes should be labeled “agent/patient nominalization” or “subject/ object nominalization” for Tibeto-Burman and others discussed below is immaterial for our immediate purposes.

 Masayoshi Shibatani

no nominalizer; instead it requires a genitive marking as in (17) above (see section 8 on the connection between genitives and nominalizations). These nominalized forms also function as arguments, as illustrated below: (19)

Northern Qiang qa [lәɣz tse-m] e: u-tɕu-a 1sg book read-nmz one.cl dir-see-1sg ‘I see one who is reading books.’

Uto-Aztecan is another language family that displays the pattern of argument nominalization similar to Western Austronesian and Tibeto-Burman. In Yaqui the distinction is made between subject nominalization (-me), object nominalization (-’u) and locative nominalization (-’Vpo), and possibly some others. (20) Yaqui (Alvarez 2007) Subject nominalization a. ju-me usi-m yeewe det-pl child-pl play ‘The children are playing.’ b. ju-me [yeewe-me] det-pl play-nmz ‘the ones playing’

(equivalent to Austronesian AF)

c. ju’u yoeme ju-me usi-m [yeewe-me] kaka-m maka-k (rc) det man det-pl child-pl play-nmz candy-pl give-perf ‘The man gave candies to the children who were playing.’ (21) Object nominalization a. inepo uka chu’u-ta tea-k 1sg det.ac dog-acc find-perf ‘I found the dog.’ b. in uka tea-ka-’u 1sg det.ac find-perf-nmz ‘what I found’

(equivalent to Austronesian pf)

c. U chu’u [in tea-ka-’u] chukuli det dog 1sg.gen find-perf-nmz black ‘The dog that I found is black.’

(rc)4

. Heath (1972:235) mentions that -’u appears “not possible to use …as a modifying RC after a head noun.’ Either the information in his source is incomplete or the use of object nominalization in RCs in Yaqui is a recent innovation.

(22)

Elements of complex structures, where recursion isn’t 

Locative nominalization Wa kari [nim bo’o-pea-’apo] ujyooli (equivalent to Austronesian lf rc) dem house 1sg.gen sleep-des-nmz pretty ‘That house where I want to sleep is pretty.’

The following are examples in which argument-nominalized forms function as an argument and as a nominal predicate paralleling some Austronesian and Northern Qiang examples above. (23) Yaqui a. [U nim pu’akta-’u] bette det 1sg.gen bear-nmz be.heavy ‘What I bear is heavy.’ ‘My burden is heavy.’ b. Jabesa [wa jiosam noktua-me] who dem book read-nmz ‘Who is the one that read that book?’ Turkish makes a distinction between subject nominalization and object nominalization in terms of different participial forms of verbs. Present participle ending -en marks subject nominalization, while in the case of (one of) the future participle(s), the nominalized form is identical with the base form. (24) Turkish (Lewis 1967:158ff) Subject nominalization a. bekliy-en-ler wait-ptcpl-pl ‘those who are waiting’ a′. [bekliy-en] misafir-ler wait-ptcpl guest-pl ‘guests who are waiting’ b. haber gelecek news will.come ‘The news will come.’ b′. gelecek ‘who/which will come, the future’ b′′. [gelecek] haber will.come news ‘news which will come’

(appositive rc)

(future sentence)

(subject nominalization) (appositive rc)

 Masayoshi Shibatani

(25) Object nominalization (-dik, -cek plus a personal suffix) a. bir tanı-dığ-ım one know-p.ptcpl-1sg ‘one I know, an acquaintance of mine’ b. [tanı-dığ-ım] bir adam know-p.ptcpl-1sg one man ‘a man I know’ (‘a man characterized-by-my knowing’)

6. The “subjects-only” constraint as an Austronesian epiphenomenon As is clear from the above exposition, many languages from different language families seem to have a relative clause formation similar to the Austronesian rc pattern, all making use of nominalized forms juxtaposed to the head nominal. Despite this similarity, none of the specialists of Tibetan or Uto-Aztecan languages speaks of the “subjects-only” constraint similar to the one noted by Keenan and Comrie (1977) on the Austronesian relativization. Ross (1995:729–730), while opting for the less charged term “pivot” than “subject”, also tells us that “in a PAN [Proto Austronesian] relative clause the (deleted) noun phrase coreferential with its head noun had to be its pivot.” This is a curious fact in view of the clear parallelism in the relativization pattern across these different language families. The answer to this puzzle that I offer is that the “subjects-only” constraint is an epiphenomenon seen only in Austronesian, where predicate formation in Proto Austronesian also involved argument nominalizations. According to Starosta, Pawley and Reid’s (1982/83) hypothesis, modern Austronesian clause structures evolved from equational predicate-topic constructions, where an arugment-nominalized form functioned as a nominal predicate, as illustrated by the Tagalog examples below: (26) Tagalog af NOMINAL PREDICATE + TOPIC a. [H�um�i-hiwa ng=karne] + [ang lalaki] red�AF�-cut gen=meat top man ‘one cutting meat’ ‘the man’ ‘The man is the one cutting meat.’ → ‘The man cut meat.’ (AF construction) b.

PF NOMINAL PREDICATE + TOPIC [Hi-hiwa-in ng=lalaki] + [ang=karne] red-cut-pf gen=man top=meat ‘one the man will cut’ ‘the meat’ ‘The meat is the one the man will cut’ → ‘The man will cut the meat.’ (PF construction)

Elements of complex structures, where recursion isn’t 

Starosta, Pawley and Reid’s (1982/83) account posits a reanalysis of equational predicatetopic construction into more tightly integrated structures in which the topic nominals have been reanalyzed as arguments of the verb of a predicate nominal, thereby creating a situation where the integrated topic nominal is understood to trigger the focus marking in the verb.5 Thus, argument-nominalized forms with a gap in Austronesian are involved in both relative clause formation and in the formation of one-place predicates, and it is this dual function of Austronesian nominalized expressions that engenders the “subjects-only” effect, if relative clause formation is believed to involve a full sentence as a modifier. It is still possible to talk about the grammatical relation of the gap in the nominalized form juxtaposed to a head nominal, though it is presumptuous to do so since the structures of nominalized forms are different from those of sentences and the grammatical status of the arguments in nominalized expressions is not entirely clear at this stage of research; e.g., is the agentive nominal marked genitive in a nominalized form (see (27b) below) really a subject like the nominative subject in an independent clause? One might characterize the following relative clause constructions in Yaqui as cases of subject relativization and object relativization on the basis of the presumed grammatical roles of the gaps in the nominalized form. (27) Yaqui Subject relativization a. ju’u yoeme ju-me usi-m [Ø yeewe-me] kaka-m maka-k det man det-pl child-pl play-nmz candy-pl give-perf ‘The man gave candies to the children who were playing.’ Object relativization b. U chu’u [in Ø tea-ka-’u] chukuli det dog 1sg.gen find-perf-nmz black ‘The dog that I found is black.’ One could also describe Austronesian relativization patterns in the same way, but as soon as one did so, the “subjects-only” constraint would disappear. In (28a) and (28b), gaps occur where a subject and an object are expected, and in (28b) what looks like a subject occurs in the genitive form, as agentive nominals generally do in nominalized forms as in Yaqui (above), Turkish, Japanese, and many others.

. See Naylor (1973) and Himmelmann (1991) for the analyses of Tagalog sentence structures in terms of equational predicate-topic constructions in which nominalizations function as a nominal predicate.

 Masayoshi Shibatani

(28) Tagalog Subject relativization? a. mga bata-ng [nag-la-laro Ø] pl child-lin AF.prf-red-play ‘children who are playing’ Object relativization? b. aso-ng [na-kita=ko Ø] dog-lin pf.perf-see 1sg.gen ‘dog that I saw’

7. Wh-relatives and their ilk Perhaps the idea that relative clauses are full clauses or sentences comes from the observation on relative clauses in English and other European languages where interrogative pronouns and other forms are used as relative pronouns standing for the gap in a relative clause, thereby insuring that all the clausal arguments are somehow preserved despite replacement in form and positional displacement of their occurrence. In this regard Bolivian Quechua provides an interesting case and a good introduction to the issues surrounding wh-relatives. Like Turkish discussed above, Quechua has two types of participial form representing agent and patient nominalizations. (29) Bolivian Quechua (Bills et al. 1971:200) a. hasut’i-q whip-nmz ‘the whipper, the one who whipped someone’ b. hasut’i-sqa whip-nmz ‘the whipped one, the one whom someone whipped’ As in the other languages examined above, these nominalized forms can function as noun modifiers similar to relative clauses. (30) a. [warmi-ta hasut’i-q] runa woman-acc whip-nmz man ‘the man who beat the woman’ b. [runa(-q) hasut’i-sqa=n] warmi man(-gen) whip-nmz=3 woman ‘the woman whom the man beat’ In addition to these participial nominalization forms, Bolivian Quechua has wh-relatives as in European languages. Interestingly, however, wh-forms also function as arguments as in (31c) and (32).

Elements of complex structures, where recursion isn’t 

(31) a. b.

pichus mikhu=n? (pichus ‘who’ = pi-chus ‘person-Dubitative’) who eat=3 ‘Who ate?’ Ni-wa-y [pichus mikhu=n] tell-1obj-imp who eat=3 ‘Tell me who ate.’

c. rikhu=ni [pichus mikhu=n] see=1sg who eat=3 ‘I saw (the one) who ate it. d. rikhu=ni runa-ta [pichus mikhu=n] see=1sg man-acc who eat=3 ‘I saw the man who ate it.’ (32) rikhu=ni [imatachus Maria ruwa=n] (imatachus ‘what (acc)’ = ima-ta-chus see=1sg what Maria do=3 ‘thing-acc-Dubitative’) ‘I saw what Maria did (the thing that she did).’ It may very well be the case that these wh-nominalized forms and their use in noun modification are a calque of Spanish wh-expressions, but what is interesting is that in other dialects of Quechua, especially Peruvian varieties, they do not seem to have developed the modification function of these wh-forms. For example, Huallaga Quechua has the following wh-nominalized forms, but apparently no headed relatives making use of these forms (David Weber & Roberto Zariquiey p.c.). (33) Huallaga Quechua (Weber 1983:62) a. [ima-wan wallpa-ta wanych-shayki-ta-pis] apamu-y what-com chicken-acc kill-sub-acc-indef bring-imp ‘Bring me whatever you killed the chicken with.’ b. rika-y may-pa away-shayki-ta-pis look-imp where-gen go-sub-acc-indef ‘Look wherever it is that you may be going.’ English wh-relatives and wh-nominalized forms (commonly known as “free relatives”) show almost complementary distribution, except for the adverbial forms where, why and how, where the two uses overlap. That is, wh-forms usable as relative clauses are not usable as free relatives, and free relatives are not generally usable as noun modifiers, as seen in the following pattern. (34)

a. b. c. d.

I saw the man who/*what won the first prize. I read the book which/*what you recommended to me. I will meet the man who/*whoever comes tomorrow. I read what/*which you recommended to me.

 Masayoshi Shibatani

e. I will meet whoever/*who comes tomorrow. f. I will visit the place where you live. g. I will visit where you live. The pattern like this has prevented us from considering free relatives as nominalized expressions in which interrogative pronouns simply stand for the gaps created by the argument nominalization process and from analyzing wh-relatives as simply making use of these wh-nominalized forms as modifiers. If we look at other European languages, such a possibility emerges more clearly as we see a great deal of overlap between wh-relatives and free relatives or between other types of relative clause and their independent nominal use outside the relativization context. According to Adams (1972:9), Ancient Greek relative clauses are traditionally divided into two types according to their use: (a) “adjectival (if the clause modifies another noun in some way)” or (b) “substantival (if the clause modifies itself)”. In traditional grammar, the term “substantive” is used to refer to nouns and their equivalents, and the examples below show that wh-forms in Ancient Greek are no different in function from those nominalized expressions examined above. (35) Ancient Greek (Adams 1972:9, 13) ́ a. [hòn gàr theoì philoûsin], apothnēiskei ‘whom the god loves, dies young’ b. [hós ou lambánei tòn stauròn autoû kaì akoloutheî opíso mou] ouk éstin mou áksios ‘(He) who does not take up his cross and follow me is not worthy of me’ Just like the nominalized forms studied above, these Ancient Greek forms are used to modify a noun, forming relative clause constructions as below: (36) Ancient Greek (Adams 1972:9, 14) a. Lázaros, [hòn égeiren ek nekrôn Iēsoûs] ‘Lazarus, whom Jesus had raised from the dead’ b. Teûkros, [hòs áristos Akhaiôn] ‘Teukros, the best of the Achaeans’ Latin is similar to Greek in that wh-nominalized forms can function both as arguments and noun modifiers as seen below: (37) Latin (Ehrenkranz & Hirschland 1972:24, 28) a. [qui mentiri solet], peierare consuevit ‘Whoever is in the habit of lying, is accustomed to swear falsely’ b. at sunt [qui dicant] ‘but there are those who might say’ (38) Latin (Gildersleeve & Lodge 1895:395, 396) a. Iūsta glōria, [qui est frūctus virtūtis] ‘real glory, which is the fruit of virtue’

Elements of complex structures, where recursion isn’t 

b. Uxor contenta est [quae bona est] ūnō virō ‘A wife who is good is contented with one husband.’ In modern Romance languages, we again see a great deal of overlap between the argument and the modification function of wh-forms. (39) Spanish (sp), Brazilian Portuguese (po), French (fr) a. Veré al hombre [que viene mañana] Verei o homem [que vem amanhã] Je verrai l’ homme [qui viendra demain] I will.see the man that comes tomorrow ‘I will see the man who comes tomorrow.’

(sp) (po) (fr)

b.

Veré al [que viene mañana] Veré a [quien viene mañana] Verei [quem vem amanhã] Je verrai (celui) [qui viendra demain] I will.see (to the) that comes tomorrow ‘I will see the one who comes tomorrow.’

(sp) (sp) (po) (fr)

(40) a.

Leeré el libro [que usted recomienda] Lerei o livro [que qual você recomenda] Je lirai le livre [que vous recommandez] ‘I will read the book that you recommend.’

(sp) (po) (fr)

b.

Leeré lo [que usted recomienda] Lerei o [que você recomenda] Je lirai ce [que vous recommandez] ‘I will read the one that you recommend.’

(sp) (po) (fr)

Closer matches between the relative clauses and the argument forms like the above cases are in fact a rarer pattern. In Slavic, nominalized forms headed by Russian kto ‘who’ and Czech kdo ‘who’ cannot be used as relative clauses, but those headed by chto/cto ‘what’ and kotoryj/kterej ‘which’ can modify a noun. (41) Russian a. (tot,) [kto vymyl ruki], mozhet nachatj jestj (that) who washed hands can start eating ‘The one who has washed his hands can start eating.’ b. *Maljchik, [kto vymyl ruki], mozhet nachatj jestj boy who washed (his) hands can start eating ‘The boy who has washed his hands can start eating.’

 Masayoshi Shibatani

(42) Czech a. (ten,) [kdo stojí tám-hle], ještě (that) who stand.3sg.pres there-part yet

ne-měl dort neg-have.3s.masc.past cake.acc

‘The one who stands over there hasn't got the cake yet.’

b. *kluk [kdo stojí tám-hle], ještě boy who stand.3sg.pres there-part yet

ne-měl dort neg-have.3sg.masc.past cake.acc

‘The boy who stands over there hasn't got the cake yet.’

(43) Russian a. tot, [kotoryj/chto stoit tam], eto drug otca that which/what stand there it friend father ‘The one who is standing there is my father’s friend.’ b. tot chelovek, [kotoryj/chto stoit tam], eto drug otca that man which/what stand there it friend father ‘That man who is standing there is my father’s friend.’ (44) Czech a. ten, [kterej/co sto-jí that.masc.nom which.sg.nom/what.sg.nom stand-3sg.pres

tám-hle], to je táty kamarád there-part it.sg.nom is dad.sg.gen buddy.sg.nom ‘The one standing there is my dad’s buddy.’

b. ten muž, [kter-ý/co that.masc.sg.nom man.sg.nom which-sg.nom/what.sg.nom

sto-jí tám-hle], je táty kamarád stand-3sg.pres there-part is dad.sg.gen buddy.sg.nom

‘The man who is standing there is my dad’s buddy.’

Germanic languages seem to use interrogative pronouns more sparingly in nominalization and relative clauses than in the other languages examined above. Danish does not use hvem ‘who’ and hvad ‘what’ in relative clauses at all, and they head nominalized expressions rather marginally.

Elements of complex structures, where recursion isn’t 

(45) Danish (the grammaticality rating courtesy of Bjarke Frellesvig) a. *[hvem står der] er min fars ven6 [Who is standing there] is my father’s friend. b. ???[hvem jeg så igår] er min fars ven [Whom I saw yesterday] is my father’s friend. c. ???[hvad er på bordet] er min fars [What is on the table] is my father’s. d. ?[hvad jeg læser nu] er meget interessant [What I am reading now] is very interesting. e. [hvad jeg læser nu] er avisen (“OK but not very natural”) [What I am reading now] is the newspaper It is significant that those wh-elements that do not head nominalized expressions here do not occur as relative pronouns (see also the Swedish forms below). What Scandinavian languages use as a filler for the gap in argument nominalization are adverbial forms som ‘as/like’ (for both subject and object nominalizations) and der ‘there’ (for subject nominalizations), and the nominalizations headed by these can function as modifiers. (46) a. Den [som/der står der] er min fars ven. def.art as/there stands there is my father’s friend ‘The one who is standing there is my father's friend.’ b. Mande-n [som/der står der] er min fars ven man-def.art as/there stands there is my father’s friend ‘The man who is standing there is my father’s friend.’ (47) a. Den [som jeg så igår] er min fars ven. def.art AS I saw yesterday is my father’s friend ‘The one whom I saw yesterday is my father’s friend.’ b. Mand-en [som jeg så igår] er min fars ven. man-def.art as I saw yesterday is my father’s friend ‘The man I saw yesterday is my father’s friend.’

. Notice that these forms, according to Bjarke Frellesvig, cannot be improved even if the definite article den precedes them.

 Masayoshi Shibatani

Swedish shows a pattern similar to Danish. Interesting is the fact that vad ‘what’ in Swedish, unlike the Danish counterpart, allows an argument nominalization, but its use as a nominal modifier appears not fully established. (48) Swedish a. Jag läser [vad ni rekommenderar] I read what you recommend ‘I read what you recommend.’ b. ?Jag läser boken [vad ni rekommenderar] I read book what you recommend ‘I read a book that you recommend.’ German divides gap fillers into two groups. Ones (the article series) used for argument nominalization and relative clauses, and those (interrogative pronouns) used primarily for the former. (49) German a. Ich empfange den, [der morgen kommt] I receive art art tomorrow comes ‘I receive the one who comes tomorrow.’ b. Ich empfange den Mann, [der morgen kommt] I receive art man art morning comes ‘I receive the man who comes tomorrow.’ (50) a. Ich lese (das), [was Sie empfehlen] I read art what you recommend ‘I read what you recommend.’ b. %Ich lese das Buch, [was Sie empfehlen] I read art book what you recommend ‘I read the book which you recommend.’ (51) a. Ich empfange, [wer (auch) morgen kommt] I receive who (also) tomorrow comes ‘I receive who(ever) comes in tomorrow.’ b. *Ich empfange den Mann, [wer morgen kommt] I receive art man who tomorrow comes ‘I receive the man who comes tomorrow.’ Apparently there is a dialectal difference over the use of the was ‘what’ relative seen in (50b). Sentences like this seem acceptable to the speakers of certain southern German dialects and Swiss German, but not to High German speakers. In any event, German, like Scandinavian and Slavic languages, shows the pattern where the relative clause

Elements of complex structures, where recursion isn’t 

usage of argument nominalizations constitutes a subset of the available argument nominalizations. That is, there are wh-type argument nominalizations that are not used as noun modifiers. Compared to her other Germanic sisters, English has developed the usage of wh-forms more extensively. Thus both who- and which-forms are used in relative clauses, whereas what-forms and wh.ever forms are used only outside the relativization context. But this division of labor seems a recent development, as older forms of English used which- and who-forms as arguments. Middle English Dictionary (2000) offers the following description about one of the uses of which: “[a]s an independent relative with indefinite or generalized force referring to a thing, an abstraction, etc., introducing noun clauses” (492) (see example (52a) below). Similarly whō was also used “[a]s an independent relative referring to a person or persons used as subj. or pred. nom. introducing noun clauses” (538), as illustrated in (52b). Setting aside the oddity of calling these wh-forms as “independent relatives” when there are no “antecedents” for them to relate to, these descriptions are revealing and consistent with our observation and analysis, especially in the recognition that these wh-forms head nominalized clauses and that they are referring expressions by themselves (see below). The use of who- and which-forms as arguments in Modern English is seen in what Huddleston and Pullum (2002:1076) call “free choice construction” (53a). And finally which-forms are also used together with that, similar to the Slavic pattern (41)–(44), in older expressions or as archaisms in Modern English, as in (53b). (52) Middle English (Middle English Dictionary) a. And [which falleÞ on Þat furste flur] schal beo Quene. b. [Who aske this] Leseth his asking trewely. (53) Modern English a. Invite [who/whom you like]. (Huddleston & Pullum 2002:1076) b. After silence, [that which comes nearest to expressing the inexpressible] is music. (Aldous Huxley Music at Night 1931) The pattern of development we see in English then is specialization of form for different functions, who- and which-forms for the modification function, and what- and wh.ever-forms for the argument function, which is consistent with the evolutionary pattern of adaptation and specialization. That is, English wh-argument nominalizations were once more widely used both as arguments and modifiers, where the latter were extensions of the former. They then have started to specialize in the specific functions ending up in the more-or-less complementary distribution pattern we saw earlier in (34).

 Masayoshi Shibatani

8. What is nominalization? When I asked a couple of experienced Mayan specialists if the K’ichee’ relative clause in (54a) was nominalized, their immediate answer was “No!” There is no nominalization morphology, and the verb form in the relative clause is finite with its arguments fully instantiated in clitic form. Indeed, the relative clause can stand as a complete sentence as in (54b).7 (54) K’ichee’ (Larsen & Norman 1979:357) a. lee ixoq lee [x-Ø-u-ch’ay lee achih] the woman rel asp-3sg.abs-3sg.erg-hit the man ‘the woman whom the man hit’ (Or ‘the woman who hit the man’) b. x-Ø-u-ch’ay lee asp-3sg.abs-3sg.erg-hit the ‘He/she hit the man.’

achih man

It took some convincing to change their opinion, as I tried with one of the Mayan specialists I spoke to, by showing that the relative clause in (54a) functions as a nominal argument paralleling a simple noun, as below: (55) K’ichee’ a. x-Ø-inw-il lee [ixoq] asp-3sg.abs-1sg.erg-see the woman ‘I saw the woman.’ b. x-Ø-inw-il lee [x-Ø-u-ch’ay lee achih] asp-3sg.abs-1sg.erg-see the asp-3sg.abs-3sg.erg-hit the man ‘I saw the one whom the man hit.’ (Or ‘I saw the one who hit the man.’) This simple anecdote shows how much linguists are (still?) preoccupied with form.8 “Nominalization” is a functional (not a morphological or formal) notion referring to creation of a referring expression. Some (e.g., Weber 1983) prefer the term

. Independent sentences and relative clauses, however, crucially differ in K’ichee’ in that while the former may contain full noun phrases identifying all cliticized arguments, the latter cannot; in a relative clause one of them must be gapped (see example (54a)) as in argument nominalizations in other languages. Thus, x-Ø-u-ch’ay lee achih lee ixoq (asp-3sg.abs-3sg.erghit the man the woman) ‘The woman hit the man’ is a possible sentence, but it cannot be used as a relative clause. . By this statement, no disrespect is intended to my colleagues in Mayan linguistics referred to above. Also, neither Tom Larsen nor Telma Kaan, whom I also consulted on K’ichee’, said that the relative clause in (54a) was not nominalized.

Elements of complex structures, where recursion isn’t 

“substantivization” as in traditional grammars of Greek and Latin. Even traditional terms “gerund” and “participle” do not refer to formal properties; “gerund” refers to words derived from verbs that function as nouns, and “participle” to the property of a word “partaking” of the nature of (or sharing the properties of) both a noun and an adjective. Indeed, there are lexical nominalizations that do not show any morphological change but which are recognizable as nominalizations; e.g., nouns drive, play, walk, etc. derived from the corresponding verbs in English. Thus, whether or not a form in question has a finite verb form is to a large extent irrelevant, just as isolating languages like Chinese do not show a difference in verb form in either independent sentences or nominalized forms. As long as such forms function as a referring expression like nouns and behave like nouns syntactically, they are nominalized, just as nouns drive, play, walk, etc. are. What we have been identifying as nominalized forms above may be termed “grammatical nominalizations” in opposition to “lexical nominalizations” of the employer- / employee-type. While lexical nominalization creates new lexical items belonging to the noun class of the language, grammatical nominalization creates new referring expressions that have no lexical status (see the following discussion on other properties of grammatical nominalizations). Grammatical nominalizations, especially those that show a clausal character, have often been considered a type of relativization and are called “headless relatives” or “free relatives”, as if they were derivatives of relative clauses. There is no basis for this other than the fact that they show formal resemblances to relative clauses (for the good reason clear from the discussion above) and the skewed perspective many linguistics have had about grammatical nominalizations, namely viewing them from the perspective of relative clauses. Many indeed attempt to derive these nominalized forms in question from relative clause constructions by deleting a head noun (see Adams (1972) on Ancient Greek and other contributions on different languages in the same volume in which Adams’s article is found, Weber (1983) on Quechua, Sneddon (1996) on Indonesian, Huang (2008) on Qiang). Matisoff (1973) also entertains such a possibility for Lahu nominalization but abandons it by noting the following: Overweighing this consideration is the simple fact that an underlying Nrh [relative head nominal] is not at all necessary to explain how these sentences work. To interpret the ve-clauses as ordinary nominalizations does not distort the meaning, and has the crucial advantage of avoiding multiplication of covert entities…Once we admit deleted Nrh’s after some ve’s, consistency would demand that we stick them in after every post-verbal ve, even in the (very numerous) cases where the only semantically possible Nrh would be an empty one like ɔ -́ cә̀ ‘thing’ or ɔ -́ lɔ ‘matter’. (Matisoff 1973:484–485)

Another favorite way of analyzing the relevant nominalized expressions as relative clauses is by positing a PRO that functions as a relative head, eschewing the deletion issue. For example,

 Masayoshi Shibatani

(56) Japanese a. [Taroo=ga tabe-ta]=no=wa… Taro=nom eat-past=nmz=top ‘the one that Taro ate is…” b. [[Taroo=ga tabe-ta](=no) pro]=wa9… (57) Spanish a. lo [que usted recomienda] art that you recommend ‘what you recommend’ b. lo [pro [que usted recomienda] art that you recommend ‘what you recommend’ Indeed, these nominalized forms are typically used in the context where an “antecedent” is available for identifying what they refer to. However, analyzing these nominalizations as relative clauses by positing a fictitious relative head or abstract category such as PRO only reveals lack of understanding of what grammatical nominalization is all about. Notice also that there are so-called free relatives that do not function as relative clauses (see (41), (42), (48), (50), (51)), which preclude the deletion analysis of these. While Matisoff ’s (1973:484) point that the purpose of nominalization (in using the particle ve in Lahu) “like an English -ing and to nominalization is…to reify a clause abstractly, without committing it to the modification of anything else in the sentence” is correct, its thrust is not strong enough. The limitations of Matisoff ’s understanding of nominalization are revealed by the weakness in his explanation (Matisoff 1972) for the connection between grammatical nominalization, relativization and genitivization, which he recognizes in a fair number of Asian languages. He attributes this pattern of sharing nominalization morphology (e.g., Lahu ve, Japanese no, Chinese de) in the three functional domains to “the nominalizing power of ve [and others]” and its connection to the subordinating function. (Matisoff 1972:251, 254) That is, Matisoff ’s explanation, together with the assumption that relative and genitive constructions involve subordination, is that these nominalization morphemes have a subordinating function and therefore are used in those constructions. Setting aside the real issue of why nominalization has a subordinating function, which Matisoff does not address, his assumption that both relative and genitive constructions involve subordination is questionable in many cases.

. See, for example, Wrona (2008). There is also a proposal that treats the nominalization marker no itself as a pronoun.

Elements of complex structures, where recursion isn’t 

In most cases studied above, relative clauses are not syntactically dependent upon their head and they can stand on their own, as shown below, although some languages require overt nominalization markers (as in Japanese no) or noun-marking articles (e.g., Spanish lo; see below) when the nominalized forms are used as arguments. (58) Lahu (Matisoff 1973:483, 484) a. [chu] ve vàʔ qhâʔ jâ lâ fat nmz pig expensive very Q ‘Is the fat pig very expensive?’ b. [chu] ve qhâʔ jâ lâ fat nmz expensive very Q ‘Is the fat one very expensive?’ (59) Pancor Ngeno-Ngené Sasak a. Loq Ali mbace buku [si beng=ku iye]=no art Ali N.read book nmz give=1 he=that ‘Ali read the book that I gave him.’ b. Loq Ali mbace [si beng=ku iye]=no art Ali N.read nmz give=1 he=that ‘Ali read what I gave him.’ (60) Yaqui a. ju’u yoeme ju-me usi-m [yeewe-me] kaka-m maka-k det man det-pl child-pl play-nmz candy-pl give-perf ‘The man gave candies to the children who were playing.’ b. ju’u yoeme ju-me [yeewe-me] kaka-m maka-k det man det-pl play-nmz candy-pl give-perf ‘The man gave candies to the ones who were playing.’ (61) Chinese a. nĭ méi yŏu [wŏ xĭhuān] de chènshān you not have I like nmz shirt ‘You don’t have a shirt that I like.’ b. nĭ méi yŏu [wŏ xĭhuān] de you not have I like nmz ‘You don’t have what I like.’ (62) Japanese a. Taroo=wa [boku=ga yatta] hon=o yonde-i-ru. Taro=top I=nom gave book=acc read-prog-pres ‘Taro is reading the book that I gave him.’

 Masayoshi Shibatani

b. Taroo=wa [boku=ga yatta]=no=o yonde-i-ru. Taro=top I=nom gave=nmz=acc read-prog-pres ‘Taro is reading what I gave him.’ (63) Spanish a. Leeré el libro [que usted recomienda]. I.will.read the book that you recommend ‘I will read the book that you recommend.’ b. Leeré lo [que usted recomienda]. I.will.read the that you recommend ‘I will read what you recommend.’ Contrary to Matisoff ’s (1972:242) assumption, these relative clauses do not seem to be “semantically subordinate to a noun head” either. To the extent that the (a) forms above entail the (b) sentences and to the extent that the understanding of these entailments does not depend on the understanding of the head noun, the relevant relative expressions here are not semantically subordinate to the head.10 These facts, both syntactic and semantic, accord better with our earlier suggestion that these constructions are basically appositive. The same argument applies to the genitive constructions that involve nominalization morphology. Matisoff (1972:242) believes that possessive structures in (64) below are semantically subordinate to a noun head, but the syntactic and semantic relations between the (a) and the (b) forms in (64)–(66) below suggest otherwise. (64) Lahu (Matisoff 1973:483) a. [yɔˆ] ve vàʔ qhâʔ jâ lâ he nmz pig expensive very Q ‘Is his pig very expensive?’ b. [yɔˆ] ve qhâʔ jâ lâ he nmz expensive very Q ‘Is his very expensive?’ (65) Chinese a. Zhè shì [wŏ] de shū nà shì [fùqīn] de this is I nmz book that is father nmz ‘This is my book, and that’s the father’s.’

. Compare these relative clauses with noun compounds where a noun modifies another noun. The sentence I bought a bookcase does not entail I bought a book.

Elements of complex structures, where recursion isn’t 

b. Zhè shì [wŏ] de nà hì [fùqīn] de this is I nmz that is father nmz ‘This is mine, and that’s the father’s.’ (66) Japanese a. Kore=wa [boku]=no hon de, are=wa [otoosan]=no da. this=top I=nmz book cop that=top father=nmz cop ‘This is my book and that is the father’s.’ b. Kore=wa [boku]=no de, are=wa [otoosan]=no da. this=top I=nmz cop that=top father=nmz cop ‘This is mine and that is the father’s.’ In other words, the genitive forms in these languages are nothing but nominalized forms alongside similar nominalized forms that can be juxtaposed to a nominal head as in relative expressions. Remember that Sasak (and other Austronesian languages in Indonesia) can nominalize demonstratives to yield forms like Puyung Meno-Mené siq ino ‘that one’ or Bahasa Indonesian form yang itu ‘that one’ with the nominalization markers siq and yang. In Lahu, Chinese, Japanese, and others, nouns can be further nominalized to give forms referring to an entity pertaining to the noun that is nominalized. This is precisely what the Japanese form Taroo=no e (Taro=nmz painting) ‘Taro’s picture’, for example, means, namely the juxtaposition of ‘what pertains to Taro’ and ‘painting’, where the latter identifies “what pertains to Taro”. The specific meaning of “pertaining to Taro” can vary. The interpretation of “possessed by Taro” is likely to be the most prevalent one, but other possibilities such as “drawn by Taro” and “Taro being the subject of ” exist.11 The proposed nominalization analysis of these genitive constructions also renders superfluous Matisoff ’s (1973:140ff) deletion analysis of the (b) form in (64); the (b) form in (64) parallels the (b) form in (58), to which Matisoff does not apply a deletion analysis. This parallelism is accounted for straightforwardly by our analysis of genitives as nominalizations. The essence of nominalization, then, is creation of a referring expression, hence its sharing this essential nominal function with nouns, which refers to a state of affairs characterized by an event denoted by the clause (event nominalization), to an entity characterized in terms of the denoted event in which it has crucial relevance (argument nominalization), or to an entity having crucial relevance to the referent of a noun (“genitive” nominalization). Grammatically nominalized forms are referring expressions

. Cf. Li and Thompson’s ( 1981:113ff) labeling of the Mandarin “genitive” de phrases as “associative phrases”, and Lewis’s (1967:164) literal translation of the Turkish personal participle RC form [kardeş-im-in bekle-diğ-i] misafir ‘the guest [for] whom my brother is/was waiting’(‘pertaining to my brother’s waiting’).

 Masayoshi Shibatani

by themselves without any head nominal or pronoun, either abstract or concrete, and they fill the syntactic noun function as arguments or predicate nominals just as lexical nominalizations of the type employer and employee are referring expressions by themselves and play the nominal syntactic role. Grammatical argument nominalizations referring to entities characterized in terms of events, however, differ from these lexical nominalizations in that they lack identifications, specifications or names; the lexical nominalizations above name the entities referred to. Thus, the English expression what I bought yesterday, for example, characterizes an entity referred to in terms of an event of my buying it yesterday, but its identity is not specified — it could be a book, a newspaper, a hamburger, an umbrella, or any other things that could be bought. The properties and their distribution of grammatical argument nominalizations follow from this process of creating a variety of new referring expressions pertaining to a limitless range of everyday events that vary in their content. First, an argument nominalization is associated with a presupposition that an event characterizing the entity referred to has taken place. The expression what I bought yesterday presupposes that I bought something yesterday. This property of an argument nominalization carries over to its relativization function such that relative clauses represent presuppositions, as the book which I bought yesterday presupposes that I bought a book yesterday. Thus, neither in the nominalized form nor in its relative clause function, is there assertion of the state of affairs denoted in the presupposition even if the form may contain a finite verb form as in the English examples here. This is an important distinction between sentences and nominalized clauses/forms, and which allows the latter to be embedded or subordinated into a main assertive clause (see below). Since argument nominalizations are typically created for the nonce, they are often indefinite, and this accounts for the use of indefinite interrogative pronouns such as what and who as a place holder for the gap created by the nominalization process. Though many languages require such nominalized forms to be preceded by an article or demonstrative, such an article, even if a definite one is used, does not necessarily definitize what follows. This kind of article is becoming grammaticalized and is functioning much like the nominalization markers no (Japanese) and de (Chinese) seen above. For example, the following Spanish nominalized forms headed by the definite articles do not refer to definite referents identifiable to the hearer; in (b) and (c), with the subjunctive verb forms, they are not even specific — the speaker has no specific referent in mind (see the additional relevant use of lo marking what appear to be nominalized in (67d) provided by Albert Alvarez). (67) a. María está leyendo su libro y yo voy a leer Maria is reading her book and I fut to read.inf

lo art

[que Juan me dio]. that Juan me gave

‘Maria is reading her book and I am going to read what Juan gave me.’

Elements of complex structures, where recursion isn’t 

b. La [que hable inglés] ganará el certamen. art that speak.3sg.sbjnct English win.3sg.fut art contest ‘The (female) one who speaks English will win the contest.’ c. El [que diga que es mentira] es el asesino. art that say.3sg.sbjnct that is false is art killer ‘The (male) one who says that it is false is the killer.’ d. Voy a leer lo indispensable/de Cervantes/elegido por Juan. fut to read.inf art indispensable/of Cervantes/chosen by Juan ‘I am going to read the indispensable one/the one from Cervantes/the one chosen by Juan.’ Because the identity of what are referred to by argument nominalizations is unspecified, their distribution is constrained in such a way that they typically occur in those contexts where the entity identification is provided or sought: in (a) relative clause constructions, where the head supplies the identification, (b) wh-questions that seek the identity of the entity referred to in the equation format, (c) cleft-type constructions, in which the entity identification is made in the equation format, (d) where the entity identification can be made from the context (see also example (65) and (66) above), or (e) the entity referred to is generic (see the earlier Greek and Latin examples in (35) and (37)). (68) Puyung Meno-Mené Sasak a. dengan [saq nyenke=n tokol leq bucu]=no man nmz prog=3sg sit loc corner=the ‘the man who is sitting in the corner’ b. Sai [saq nyenke=n tokol leq bucu]=no who nmz prog=3sg sit loc corner=the ‘Who is the one sitting in the corner?’ c. [saq nyenke=n tokol leq bucu]=no amaq=k nmz prog=3sg sit loc corner=the father=1sg ‘The one sitting in the corner is my father.’ d. Gitaq [saq nyenke=n tokol leq bucu]=no look nmz prog=3sg sit loc corner=the ‘Look at the one sitting in the corner’ Among these constructions, the wh-question (68b) and what looks like a pseudocleft construction (68c) are often analyzed as cases of extraction in analogy with the extraction analysis of relative clauses, where the wh-element and the identifier nominal in the pseudo-cleft are somehow extracted from the nominalized construction

 Masayoshi Shibatani

(see, for example, Larsen & Norman 1979). There is no real motivation for such an analysis. These constructions are the same as simple equational constructions of the following type, which no one would analyze as involving extraction of an element. (69) Puyung Meno-Mené Sasak a. Sai ino? who that ‘Who is that?’ b. Kamu amaq=k you father-1sg ‘You are my father.’ The only difference between these simple equational sentences and the ones involving nominalized forms is that the latter have presuppositions associated with nominalization, such that (68b–c) presuppose that there is someone sitting in the corner (see above). While in Western Austronesian languages questions involving nominalized forms as in (68b) are normal question forms, some languages have both simple and complex wh-question formation, the latter of which involves nominalized forms. For example, in Yaqui either (70a) or (70b) below can be used, the difference being that the latter has a presupposition associated with the nominalized form that someone read that book. (70) Yaqui a. Jabesa jiosam noktua who book read “Who read the book?” b. Jabesa [wa jiosam noktua-me] who that book read-nmz ‘Who is the one that read that book?’

9. Conclusion: Sentence and clause I opened this paper by noting that the distinction between sentences and clauses is not made in the Chomskyan recursive phrase structure rules, which imply that sentences may recur in clause internal positions. Though an intuitive understanding of the distinction between the two seems to be there, linguists have generally tended either to gloss over the distinction or to be confused about the relationship between the formal finiteness features characterizing sentences and the predication function they perform. This confusion is understandable because certain embedded clauses do display some formal finiteness features such as tense marking and agreement features in the verb. While

Elements of complex structures, where recursion isn’t 

the distribution of formal finiteness features in different types of subordinate clauses deserves close study, finite subordinate clauses are by no means functionally finite — and are accordingly not sentences — in that they do not assert, order, warn, promise, or express the speaker’s ideas/desires/surprises, lacking this central function of sentential predication. Formal finiteness features such as tense inflection in a main clause mark elements that support assertion (e.g., by locating a proposition at a specific time) but their presence in a subordinate clause does not mean that an assertion, for example, is made, as is apparent in the indicative relative clause. As pointed out above, relative clauses, whether they are subordinated to the head or appositive, do not make an assertion, as is clear from the well-known negation test: I didn’t read the book which John recommended to me does not negate the presupposition that John recommended something to me. Likewise, the Latin subjunctive found in a nominalized form like Habeō [quae velim] ‘I have what I should like’ does not express the speaker’s opinion “I should like/wish something” as the subjunctive sentential form Velim does.12 Thus, even if relative clauses and other types of modification elements are finite and clausal, they are never sentences. Indeed, the fact that a single sentence makes only one assertion seems to reflect a strong cognitive constraint on linguistic structures, and thus failure to make a clear distinction between clauses and sentences misses this very important aspect of human linguistic ability.13 While our argument is that many of the so-called relative clauses are not full clauses, let alone full sentences, and while it is not easy to characterize what is and what is not a clause (e.g., is a nominalization based on a transitive verb with a genitive agent and a gapped patient such as Japanese [Taroo=no Ø katta]=no ‘what Taro bought’ a clause?), we can still talk about the degree of nominalization in terms of how similar the internal syntax of a given nominalization is to a full clause with a finite verb and the full array of its arguments realized.14 However, a categorical division between participial nominalizations and those containing a finite verb, for example, is not possible.15 Participial forms may inflect for tense, as in Palauan, where participles are a separate form class inflecting for the past and the future tense (e.g., a ʔelat el ngikel (art smoked LIN fish) ‘smoked fish’; a lelukl el babier (art to.be.read LIN

. Significantly, even in Latin, relative clauses do not admit the imperative mood. . See Chafe (1987) and Hurford (2003) on the cognitive limitations on a sentence. . See Lehmann (1986) for a relevant discussion and his attempt to identify typological correlates of the degree of nominalization. . See footnote 2 for proposals to distinguish between, for example, gerundive modifiers and relative clauses structurally. These proposals suffer from their failure to recognize the possibility of nominalizations with finite verb forms. See Noonan (2007) and Koptjevskaja1-Tamm (1993) on this and related issues in nominalizations.

 Masayoshi Shibatani

book) ‘a/the book to be read’ (Foley 1976:16)). (Also see the Turkish future participle in (24) above.) Participial nominalizations may contain a nominative or an accusative argument, while others turn them (optionally) into genitive (see Quechua examples (30)). Some nominalizations with a finite verb contain a subject in nominative form, while others turn such an argument into genitive (optionally in some case; see the Japanese example immediately above in this paragraph). In still some other forms, e.g., so-called gerunds, arguments may be missing altogether (e.g., [Reading] keeps your brain going). Nominalizations of different internal syntax may coexist in a single language, and it is legitimate to ask what such formal differences correlate with. Gołąb (1972:30–31) offers a ready answer for this as follows: The question is whether the two categories under consideration, relative clauses and corresponding participles, are functionally identical…The replacement by a participle of a relative clause with a predicative verb preserves the basic external function of the relative clause with respect to its governing noun or pronoun, i.e., the function of an “adnominal” modifier. This replacement, however, deprives the relative clause of grammatical predication, thus destroying its status as a sentence. This ultimately means that a process presented primarily as something concrete, taking place in time and space, after the elimination of the relative clause, is conceived as something abstract, beyond time and space. Thus nominalization of a verbal clause deprives it of very important semantic features connected with the “verbality” itself. So, relative clauses and corresponding participles…are not identical semantically, and it is easy to see why many languages contain in their systems two different ways involving different semantic connotations. (Emphasis added)

Everything that Gołąb says above is correct, except he makes the cardinal mistake of confusing sentences and clauses; relative clauses are not sentences, as argued above. They are at best nominalized clauses lacking the predication function of making assertion, commanding, etc. It is this lack of the most essential function of predication that makes nominalized expressions incorporable into a sentence allowing them to function as modifying elements in various constructions such as relative clauses, noun and verb complements, and other types of subordinate modifier.16 This is the power of nominalization and its connection to the subordination function (Matisoff 1972:251–254).17

. See (10) for the variety of context in which nominalized forms are used in Pancor Ngeno-Ngené Sasak. . After completing this article, I came across the following passage in the revised version of the paper on lexical nominalization by Comrie and Thomson (2007:379), where they acknowledge the connection between nominalization and relativization: “Hence our claim that in certain languages relativization is indistinct from nominalization”. My claim of this paper is much stronger than this in that I claim that in a large number of languages, if not most, around the globe, nominalization is the basis for relativization.

Elements of complex structures, where recursion isn’t 

References Adams, D.Q. 1972. Ancient Greek. In The Chicago Which Hunt: Papers from the Relative Clause Festival, P.M. Peranteau, J.N. Levi & G.C. Phares (Eds), 9–22. Chicago IL: Chicago Linguistic Society. Alvarez, A. 2007. Relative clauses and nominalization in Yaqui. Paper presented to the Seminario de Complejidad Sintáctica, Universidad de Sonora, Hermosillo, México. November, 2007. Bills, G.D., Bernardo Vallejo C. & Troike, R.C. 1969. An Introduction to Spoken Bolivian Quechua. Austin TX: University of Texas Press. Chafe, W. 1987. Cognitive constraints on information flow. In Coherence and Grounding in Discourse, R. Tomlin (Ed.), 21–51. Amsterdam: John Benjamins. Comrie, B. 1998. Rethinking the typology of relative clauses. Language Design 1(1): 59–86. Comrie, B. & Keenan, E.L. 1979. Noun phrase accessibility revisited. Language 55(3): 649–664. Comrie, B. & Horie, K. 1995. Complement clauses versus relative clauses: Some Khmer evidence. In Discourse Grammar and Typology: Papers in Honor of John W.M. Verhaar, W. Abraham, T. Givón & S.A. Thompson (Eds), 65–75. Amsterdam: John Benjamins. Comrie, B. & S.A. Thompson. 2007. Lexical nominalization. In Language Typology and Syntactic Description. Vol. 3, T. Shopen (Ed.), 334–381. Cambridge: Cambridge University Press. Ehrenkranz, J. & E.C. Hirschland. 1972. Latin. In The Chicago Which Hunt: Papers from the Relative Clause Festival, P.M. Peranteau, J.N. Levi & G.C. Phares (Eds), 23–29. Chicago IL: Chicago Linguistic Society. Foley, W.A. 1976. Comparative Syntax in Austronesian. Ph.D. dissertation, University of California, Berkeley. Gildersleeve, B.L. & Lodge, G. 1895. Gildersleeve’s Latin Grammar, 3rd Edn. Hampshire: Macmillan. Gołąb, Z. 1972. The relative clauses in Slavic, Part 1: Common Slavic and North Slavic. In The Chicago Which Hunt: Papers from the Relative Clause Festival, P.M. Peranteau, J.N. Levi & G.C. Phares (Eds), 30–39. Chicago IL: Chicago Linguistic Society. Hauser, M.D., Chomsky, N. & Fitch, W.T. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298(22): 1569–1579. Heath, J. 1972. Uto-Aztecan relative clauses. In The Chicago Which Hunt: Papers from the Relative Clause Festival, P.M. Peranteau, J.N. Levi & G.C. Phares (Eds), 230–245. Chicago IL: Chicago Linguistic Society. Heine, B. & Kuteva, T. 2007. The Genesis of Grammar: A Reconstruction. Oxford: OUP. Himmelmann, N.P. 1991. “The Philippine challenge to universal grammar”. Arbeitspapier Nr. 15 (Neue Folge) Institut für Sprachwissenschaft, Universität zu Köln. Hopper, P. & Traugott, E.C. 1993. Grammaticalization. Cambridge: CUP. Huang, C. 2008. Relativization in Qiang. Language and Linguistics 9(4): 735–768. Huang, L. 2002. A Study of Mayrinax Syntax. Taipei: Crain. Huddleston, R. & Pullum, G.K. 2002. The Cambridge Grammar of the English Language. Cambridge: CUP. Hurford, J. 2003. The neural basis of predicate-argument structure. Behavioral and Brain Sciences 26(3): 261–283. Keenan, E.L. 1985. Relative clauses. In Language Typology and Syntactic Description, Vol. 2: Complex Constructions, T. Shopen (Ed.), 141–170. Cambridge: CUP. Keenan, E.L. & Comrie, B. 1977. Noun phrase accessibility and universal grammar. Linguistic Inquiry 8(1): 63–99.

 Masayoshi Shibatani Koptjevskaja-Tamm, M. 1993. Nominalizations. London: Routledge. LaPolla, R.J. with Huang, C. 2003. A Grammar of Qiang with Annotated Texts and Glossary. Berlin: Mouton de Gruyter. Larsen, T.W. & Norman, W.N. 1979. Correlates of ergativity in Mayan grammar. In Ergativity: Toward a Theory of Grammatical Relations, F. Plank (Ed.), 347–370. London: Academic Press. Lehmann, C. 1986. On the typology of relative clauses. Linguistics 24: 663–680. Lewis, G.L. 1967. Turkish Grammar. Oxford: Clarendon Press. Li, C.N. & Thompson, S.A. 1981. Mandarin Chinese: A Functional Reference Grammar. Berkeley CA: University of California Press. Middle English Dictionary. 2000. Part W4, Part W5. Ann Arbor MI: University of Michigan Press. Noonan, M. 2007. Complementation. In Language Typology and Syntactic Description, 2nd Edn. Vol. 2, T. Shopen (Ed.), 52–150. Cambridge: CUP. Matisoff, J.A. 1972. Lahu nominalization, relativization, and genitivization. In Syntax and Semantics, Vol. 1, J.P. Kimball (Ed.), 237–257. New York NY: Academic Press. Matisoff, J.A. 1973. The Grammar of Lahu. Berkeley CA: University of California Press. Naylor, P.B. 1973. Topic, focus, and emphasis in the Tagalog verb clause. Ph.D. dissertation, University of Michigan. Ross, M. 1995. Reconstructing Proto-Austronesian verbal morphology: Evidence from Taiwan. In Austronesian Studies Relating to Taiwan, P. Jen-kuei Li, Cheng-hwa Tsang, Ying-kuei Huang, Dah-an Ho and Chiu-yu Tseng (Eds), 727–791.Taipei: Institute of History and Philology, Academia Sinica. Shibatani, M. 2008. Relativization in Sasak and Sumbawa, Eastern Indonesia. Language and Lingusitics 9(4): 865–916. Sneddon, J.N. 1996. Indonesian: A Comprehensive Grammar. New York NY: Routledge. Starosta, S., Pawley, A.K. & Reid, L.A. 1982/83. The evolution of focus in Austronesian. In Papers from the Third International Conference on Austronesian Linguistics, Vol. 2: Tracking the Travellers [Pacific Linguistics C-75], 145–170. Canberra: The Australian National University. Weber, J.D. 1983. Relativization and Nominalized Clauses in Huallaga (Huanuco) Quechua. Berkeley CA: University of California Press. Wrona, J. 2008. The nominal and adnominal forms in Old Japanese: Consequences for a reconstruction of pre-Old Japanese syntax. In Proto-Japanese. B. Frellesvig & J.B. Whitman (Eds), Amsterdam: John Benjamins. 199–221.

Nominalization and the origin of subordination Guy Deutscher This paper argues that the paths portrayed in recent literature as the genesis of subordination are only superficial rearrangements of existing subordination, while the real syntactic-cognitive underpinnings of subordination are overlooked. (Derivational) nominalization, the ability to derive a noun from a verb, is shown as the core element in the channel of ‘expansion’, and may also be behind the genesis of relative clauses that are claimed to arise through ‘integration’. And yet, the origins of nominalization are little researched and understood, and thus accounts of the genesis of subordination are robbed of much of their explanatory power. One way is suggested to account for the genesis of nominalization without already presupposing it, based on back-formation from the process of verbalization.

There is a Jewish story about a man who is desperately searching for his keys on the pavement under a street light. When a passer-by offers to help and asks where exactly the keys were lost, the man explains that they fell out of his pocket somewhere further down the road. “So why are you searching for them here?” asks the passer by. “Because it’s much easier to search under the light.” This paper argues that when seeking the origin of subordination, we also run the risk of looking for it under the light, in the easily visible, but relatively superficial elements of the process. I argue that some of the paths which have been portrayed in the literature as the genesis of subordination may actually only be rearrangements of already existing subordinate structures, while the real syntacticcognitive underpinnings of the genesis of subordination have been overlooked. In particular, I suggest that (derivational) nominalization is an unsung hero in the story of subordination. The ability of a language to derive a noun from a verb, that is, to reify a verbal predicate and to present it as a nominal argument or modifier, is at the core of subordination. And yet, the origins of nominalization are little researched and little understood, and thus the standard accounts of the genesis of subordination are robbed of much of their explanatory power. (It is important to stress at the outset that by ‘nominalization’, I refer throughout this paper only to the derivational process of turning a verb into a nominal form such as action-nominal, masdar, participle, infinitive, gerund, and the like. Thus, ‘nominalization’ here will not refer to ‘syntactic nominalization’, the

 Guy Deutscher

process of repackaging a whole clause as a noun, because this is already a fully-fledged strategy of subordination.) In their recent monograph on the Genesis of Grammar, Heine and Kuteva (2007), following Givón (this volume), suggest a binary typology for the paths through which subordinate clauses arise. They term the two main channels expansion (of a nominalized argument to a clause) and integration (of two independent clauses into one). I will consider both these channels, and suggest that they both run the risk of explaining the ‘rise’ of subordination by presupposing what they aspire to explain. The accounts of expansion take as their starting point nominalized structures that to all intents and purposes are already subordinated, and many examples of ‘integration’ also merely describe the rearrangement of already subordinate structures.

1. The role of nominalization in the path of ‘expansion’ 1.1 Nominalization presupposed Discussions of the emergence of subordination through expansion usually take a nominalized verb as given, and concentrate only on the paths by which such a nominalized form can develop into a more clause-like subordinate structure. Harris and Campbell (1995:310–13), for instance, suggest that the ultimate origin of subordination can be understood as the extension of structures such as ‘I saw the dancing girl’ to subordinate clauses such as ‘I saw the girl who had danced’. The origin of ‘I saw the dancing girl’ is not deemed to require explanation, and their conclusion is that the ‘first introduction of subordination’ in language can be explained by the path which leads from “dancing girl” to “a girl who was dancing” [my emphasis]’. Heine and Kuteva (2007) likewise assert that the expansion of nominalized verbs can explain the genesis of subordination. They argue (2007:214) that the scenario of expansion entails a ‘strong claim, namely that clause subordination is historically derived from non-subordinate sentences’, implying again that the initial stage in the process, a nominalized verb in embedded position, is not yet subordinate and hence requires no explanation. Heine (this volume) outlines five stages in the process of expansion from NP to fully verbal clauses. Stage 0 consists of a simple noun in an embedded position of argument or adjunct, as in (1a) below. Stage 1 is ‘an extended noun stage’, in which a non-finite verb ‘typically in a nominalized, an infinitival, or an participial form’ appears in an embedded position as argument or adjunct, but still has the internal structure of a noun phrase, as in (1b). As the embedded clauses increasingly acquire verbal properties, they proceed through stages 2 and 3, and finally to Stage 4, subordinate clauses that are indistinguishable in their morpho-syntax from finite main clauses, as in (1c).

(1)

Nominalization and the origin of subordination 

a. Stage 0: [The aardvark] drew international attention b. Stage 1: [Algernon’s shooting of the aardvark] drew international attention … c. Stage 4: [that Algernon shot the aardvark] drew international attention

Heine demonstrates how languages proceed gradually from Stage 1 to Stage 4. However, we cannot claim to have explained the rise of subordination, unless we also understand the change from Stage 0 to Stage 1. If we want to understand how subordination arises from structures that are genuinely cognitively simpler, then Stage 1 cannot just be assumed as an axiom, for in terms of cognitive complexity, Stage 1 is subordination in all but name. If a language has the ability to create a phrase such as ‘Algernon’s shooting of the aardvark’, and to embed this phrase in another clause, then it already has the means of squeezing a whole proposition, with predicate, arguments and all, as an argument of another predicate. Moreover, the pattern in (1b) even allows recursion: ‘Algernon’s financing of the filming of the shooting of the aardvark drew international attention’, and so on. In fact, if there is any real cognitive leap in the genesis of subordination, it is between Stage 0 and Stage 1. Once that leap has been made, once a verb has been packaged into a tight nominal wrapping and in this way embedded in a higher clause, the further expansion is a fairly natural sequel: it is only natural that speakers would try to let the verb expand from its restrictive pressure-packaging, regain some of the flexibility of an independent verb, and thus achieve the ability to convey the additional information that an independent verb typically supplies (arguments, TMA, and so on). As Heine shows, this expansion is achieved through small-step morphosyntactic rearrangements. The only cognitively challenging step in the process, therefore, is the genesis of the nominalized verb in embedded position, that is, the process that leads from Stage 0 to Stage 1. And yet, exactly this crucial step has so far been presupposed in discussions of expansion. The most significant part of the process has thus been left unaccounted for, and thus the accounts of expansion have so far failed to explain the origin of subordination.

1.2 The elusive origin of nominalization How much do we actually know about the origin of (derivational) nominalization and of nominalizing markers? The genesis of grammatical markers has been the subject of intense study in the last few decades. And yet, the rich literature on grammaticalization is strangely silent on the origin of nominalizing markers. Hopper and Traugott’s Grammaticalization (2003), for instance, does not mention ‘nominalization’ in the index. Heine and Kuteva’s Lexicon of Grammaticalization (2002) has no mention of the source of nominalizing morphemes. Cross linguistic surveys of nominalization, such as Comrie and Thompson (2007) and Koptjevskaja-Tamm (1993, 2003)

 Guy Deutscher

are wholly synchronic in orientation, and do not explore the origin of nominalizing markers either. At this point, one may raise the following objection: a lot of research has been done on the grammaticalization of case markers into markers on dependent verbs (Genetti 1991; Blake 1999, to name but a few), and in particular on the development of dative/allative markers into infinitive markers (Haspelmath 1989; Hopper & Traugott 2003:188ff., and many others). Are such processes by which case markers are extended to verbs not exactly the paths by which nominalized verbs arise? The problem is that if we look at the process in languages where verbal and nominal bases are clearly distinguished, we see that when case markers are extended to verbs, they are actually added to ‘verbal nouns’, that is, to verbal forms that are already nominalized. The infinitive stems of modern Indo-European languages go back to such old verbal nouns (Szemerényi 1996:324–7). The dative adpositions that created the infinitives in modern European languages were added not to finite verbs, but to action-nominals and other nominalized forms. So the case markers did not create nominalization in the first place, they merely renewed nominal morphology on verbs that were already nominalized. Similarly, in Akkadian, the dative/ allative preposition ana turned into an infinitive marker (Deutscher 2000:128–9), but the preposition was added to a verbal form that had already been nominalized. We thus see that the grammaticalization of dative markers into infinitive markers does not explain the origin of nominalization. Rather, it presupposes nominalization. More generally, therefore, it is plausible that the extension to verbs of other nominal markers, such as classifiers (Aikhenvald 2002:220) or demonstratives (Greenberg 1991), may likewise only reflect the extension of nominal morphology to already nominalized verbs.1 A potentially more promising source for nominalizing morphemes appears, at first sight, to be compounding structures, as discussed primarily in Tibeto-Burman languages and other languages from East Asia (see collection of papers in http://tibeto-burman. net/nominalizationworkshop.html), and occasionally in languages elsewhere, such as Basque (Trask 1995) or Supyire (Niger-Congo) (Carlson 1994:108–16). In such languages, lexical origins for nominalizing morphemes have been suggested, especially for agent nominalizers, place-nominalizers, time-nominalizers, and instrument-nominalizers.

1. Indeed, the existence of remnants of nominal markers on finite verbs may in many cases not be the product of extension of nominal markers to finite verbs at all, but rather the result of a process by which nominalized verbs drift into the finite verb system by gradually acquiring more verbal characteristics (Givón 1994). The renewal of the finite verb systems through gerunds, participles and other verbal nouns is a common and well known process (Bybee et al. 1994). In the history of the Semitic languages, for instance, we can follow multiple waves of such ‘finitizations’ of verbal nouns (see e.g., Cohen 1984; Kouwenberg forthcoming, ch. 4), and the resulting finite forms may still carry with them remnants of their earlier nominal morphology.

Nominalization and the origin of subordination 

Not surprisingly, the origins of such nominalizing morphemes can sometimes be traced back to full nouns meaning ‘person’, ‘place’, ‘time’, ‘thing’, and so on. For instance, in Supyire, the nominalizing prefixes ya- and ka- are transparently related to the nouns yaaha and kya, both meaning ‘thing’ (Carlson 1994:112). In Tibetan, a nominalizing suffix -sa is argued to have been historically a noun meaning ‘ground’ or ‘place’ (Delancy 1986). The argument is that what started originally as N-N compounds (‘N-place’) was extended to V-N compounds (‘V-place’), such as yod-sa ‘live-place’, and then the head noun lost its independent status, and turned into a nominalizing suffix. While there is no reason to doubt the ultimate lexical origin of morphemes such as Tibetan -sa, the alleged nominalizing power of compounds suffers from the same problem as the alleged nominalizing power of case markers: one has to be wary of the possibility that the V-N compounds are formed with verbs that are already nominalized. In languages where there is a clear morphological distinction between noun and verb bases, we see that it is difficult to form V-N compounds like ‘live-place’ with a real verb. Rather, the first element in the compound is already nominalized: ‘livingplace’. So the compounding scenario may again simply presuppose what it alleges to explain. Compounds with sa and other such nouns may not be the source of nominalization, as they may rely on pre-existing nominalization of the verb (even if that is zero-nominalization). Indeed, Trask (1995) showed that the Basque nominalizing morphemes that go back to nouns (meaning ‘time’, ‘abundance’, and so on) originally appeared in compounds with verbs that had already been nominalized. Where, then, are those elusive grammaticalization paths that create the categorychanging markers V>N in the first place? The obscure origin of nominalization markers appears all the more striking when we compare it with the opposite category changing process: verbalization. The origins of verbalization markers are well known and can be quite straightforward: there are constructions in which a verbal lexical head grammaticalizes, loses its lexical status, and in so doing bestows upon the resulting construction a verbal status. An obvious example is factitive verbalizing morphemes that derive from ‘make’ verbs. The English -fy, for example, ultimately derives from the Latin verb facere ‘make’, through the grammaticalization the construction ‘make N/Adj’. The verbal head was bleached, coalesced with the noun, and the resulting construction thus gained the verbalizing function. So why do we not find in the grammaticalization literature examples of equally simple origins for V>N category-changing markers? Why do we not hear about parallels in nominalization to the straightforward genesis by of verbalizers like -fy? The answer seems to be that we do not hear about such paths because they do not exist. And the reason for that is a fundamental asymmetry between the syntactic status of the two category changing operations. The reason why the origin of verbalizers can be a simple path of grammaticalization is that there are source constructions in which a lexical verb functions as a head of a complex that includes a noun (as in the case facere > -fy).

 Guy Deutscher

But there are no source constructions with a nominal head that can grammaticalize directly into V>N category-changing morphemes. Or to be more accurate, the only source constructions that have a noun as a head and a verb under it are ones that already involve nominalization or subordination. If we are not allowed to presuppose subordination or nominalization, there is no syntactic arrangement in which a noun serves as a head of complex which contains a verb. It is no wonder, then, that it is difficult to find direct grammaticalization paths for the origin of nominalizers. Indeed, if we look at the common nominalizing suffixes of English, for instance, both those of Germanic origins (such as -ing) and Latinate origins (-tion, -age), we see that none of them has a ‘straightforward’ history of grammaticalization, but all have rather convoluted backgrounds. What seems to unite them is that they always start out as N>N (or ADJ>N) derivations, creating collective nouns or abstract nouns from simple nouns. For example the Germanic suffix -ing/ung seems originally to have been denominal. Kluge (1995, s.v. -ung) explains: ‘Letztlich liegen indogermanische k-Erweiterungen zu n-Stämmen vor, so daß das Suffix ursprünglich denominal gewesen sein muß’. Indeed, in Old English and older stages of other Germanic languages, -ing/-ung appeared more commonly with nominal bases (such as cyn-ing ‘king’, Idum-ing ‘Edomite’ (cf. Munske 1964:66ff. & Jespersen 1948:205). The extension to verbs, and the function of nominalizing them, must therefore be a later development. So we seem to be getting back to the same problem again and again: the morphemes that we see synchronically in the role of nominalizing verbs all somehow appear to go back to elements that were originally attached to nouns. In some mysterious way, such denominal morphology then migrates to verbs. But how can the migration be explained unless the verbs had already been nominalized in the first place?

1.3 Possible origin of nominalization through back-formation One mechanism that can provide a solution and account for the rise of nominalization without presupposing it is a conceptual back-formation, which derives nominalization from verbalization. For this process, we have to presuppose the ability to verbalize, that is, we have to assume that a language already has the ability to take a noun N and use it as a label for a certain action related to the noun N (e.g., ‘to skin’, ‘to water’, ‘to oil’). One could object, of course, that by presupposing verbalization we are again assuming almost what we were meant to explain. However, both syntactically and cognitively, derivational verbalization is an entirely different operation from derivational nominalization, and by no means merely a reverse of the same operation (Hopper & Thompson 1985:176). So by assuming it, we are not presupposing the explanandum.2

2. Hopper and Thompson (1985:176) argued that there is fundamental asymmetry between the cognitive difficulty of derivational verbalization and derivational nominalization, and that

Nominalization and the origin of subordination 

If we assume verbalization, then we can derive nominalization from it by reverse analogy. The history of the French nominalizer -age can be used to demonstrate the steps involved in the process. In modern French, -age is used as a nominalizer on a large number of verbs: arrivage, arrosage, chauffage, pliage, raffinage, démontage, nettoyage, and so on. But the origin of this suffix are clearly denominal. It comes from the Latin suffix -(a)ticus, which was used to form the designation ‘something that relates to N’ from a noun N: aqua-ticus ‘something relating to water’, silva-ticus (something relating to the woods/wild (Modern French sauvage), and so on. In Old French, the suffix, which had become -age after the relevant sound changes, was still commonly used in N>N derivations, for forming collective and abstract nouns from simpler nouns: visage ‘collection of traits which make up the visus (appearance), fleurage ‘ensemble of flowers’, corage (from cor ‘heart’), hommage, vasselage, frerage, orphelinage, and so on. It is not known exactly with which verbs the passage of -age to verbs started. MeyerLübke (1966:61–3) uses as a demonstration the verb auner ‘to measure by the aune’. But an equally good candidate could have been marier ‘marry’ (Deutscher 2005:249–251). The process, in any case, could have been one of back-formation. French had a noun mari ‘husband’, to which -age seems to have been added directly, giving mari-age ‘the state of being a husband’. But the noun mari also gave rise to a verb marier ‘to marry’ through a well-established pattern of verbalization. There were thus two different words in the language which both derived from the noun mari: the abstract noun mariage and the verb marier. As these two are so close in meaning, it was natural for the verb and the abstract noun to come to be linked in speakers’ minds. The role of the

verbalization is by far the easier way around. Verbalization may require the imposition of argument structure, but it requires no abstraction: the meaning of a denominal verb ‘to N’ is generally just a convenient label for any plausible activity involving the noun N in some way (to milk, to oil, etc. cf. also Clark & Clark 1979; Woodworth 1991). Denominal verbs can thus refer to simple physical actions just like ‘proper’ original verbs. Going in the other direction, however, is a more involved affair from a cognitive perspective. An ‘explosion’ is not a physical object that is somehow involved in the action of exploding, it is the action itself, presented as a thing. The reification required in the creation of such an action-nominal involves a high degree of abstraction. And indeed, action-nominals and similar deverbal nouns are often abstract nouns, and often do not have the full range of morphological properties of normal nouns. Hopper and Thompson’s claim seems to me to be not just true, but self-evidently so. In an earlier version of this paper, therefore, I suggested that if we presuppose verbalization to explain nominalization, we are not just presupposing a different operation, but also one that is cognitively simpler and thus one that is likely to have developed earlier in the evolution of language. However, the claim that verbalization is easier was met with some resistance from some participants in the Symposium on which this volume is based. A full argument in support of this claim will need to be developed elsewhere. But for our present purpose, it is sufficient to stress that verbalization is a fundamentally different process from nominalization, both in cognitive terms and in syntactic terms.

 Guy Deutscher

noun mari as the original link between the abstract noun and verb could have faded from linguistic consciousness, and so speakers could naturally assume that mariage was directly derived from the verb marier, and that the abstract noun thus denoted, not ‘the state of being a mari’, but rather ‘the state resulting from the action of marier’. This conceptual back-formation thus invested -age with a new power which it had not possessed before, to create an abstract noun from a verb. And once the link was established, the pattern could be extended to other verbs, including those that, unlike marier, had never been denominal to start with. In theory, a similar process of conceptual back-formation can also explain the rise of zero-nominalizations (see now also Post 2008.) Clearly, the argument presented here need to be developed at much greater depth. But it does at least provide one way in which nominalizing morphology could have made its way onto verbs, and a way in which morphemes which had not previously had the power to affect a V>N transformation acquired this power. The important point about such a scenario is that it does not require us to presuppose nominalization (although it does presuppose verbalization), and so it can actually explain the first emergence of nominalization in a language, and hence the first emergence of subordination.

2. Nominalization in ‘integration’ 2.1 Do relative clauses emerge through integration? The second path that Heine and Kuteva (2007:225) suggest for the emergence of subordinate clauses is the integration of two independent clauses into one. They argue in particular that relative clauses headed by what had originally been a demonstrative pronoun arose though the integration of two independent clauses. While I am not disputing that integration of two clauses can be the source of some subordinate clause (adverbial clauses and complement clauses in particular), in this section I argue that the attempts to explain the genesis of relative clauses through integration also run the risk of explaining subordination by presupposing subordination, and that in fact, nominalization may play a more important role in the genesis of relativization than what it is generally given credit for. Heine and Kuteva sketch their suggested path of integration through the ‘demonstrative channel’ as follows: (2) From [S1+S2] juxtaposition to S1[S2] relativization: a. There is the car; that (one) I like b. There is a car [that I like] In their scenario, two initially independent clauses, one of which headed by a demonstrative, are integrated somehow into a main clause and relative clause. However, when

Nominalization and the origin of subordination 

one examines the actual examples adduced for this path of integration by Heine and Kuteva (as well as by Hopper & Traugott 2003:196ff., and by Givón, this volume), one finds that the process that they attest to is rather different, and should be sketched as in (3): (3) a. There is the car; that (one)[I like] (i.e., ‘there is a car, that one which I like’) b. There is a car [that I like] The process which creates relative clauses like (3b) is thus the integration of a main clause with an already existing relative clause, one that had originally been headed by a demonstrative. In this process, the erstwhile head of a relative clause loses its independence and becomes a mere relativizer. But this process does not account for the first emergence of relativization. It is simply the rearrangement of already subordinate structures. In the following section, I would like to demonstrate the difference between the alleged path of integration in (2) and what I argue to be the actual path in (3), using examples from Akkadian, because this language not only neatly shows the difference between the paths, but also gives a good indication of the actual ultimate origin of relative clauses.

2.2 Akkadian relative clauses Akkadian is a Semitic language which was spoken in ancient Mesopotamia, and is attested in writing over roughly two millennia, starting around 2500BC. The earliest attested period, from 2500BC to 2000BC, is conventionally called ‘Old Akkadian’, and it will be the main subject of discussion here. The main productive type of relative clauses in Akkadian was marked by an element that must originally have been a demonstrative pronoun (and whose cognates serve as demonstratives in other Semitic languages). The forms of this demonstrative in Old Akkadian are given in (4): (4) masc.sg masc.pl fem

nom šu šūt šāt

acc ša šūt šāt

gen ši šūti šāti

The relative marker agreed with the head noun in case, gender, and number, as in the examples below: (5) Šarru-kīn šar māt-im [šu Enlil māḫir-a lā Sargon king.of land-gen [rel(nom.m.sg) Enlil rival-acc not iddin-u-šum] he.gave-sub-to him] ‘Sargon, king of the land, that Enlil has not given him a rival … ’ (Sargon, king of the land, to whom (the god) Enlil has given no rival)

 Guy Deutscher

(6) eql-am [ša … nītiq-u] lišqi’ū field-acc [rel(acc.m.sg) we.passed-sub] they.should.water ‘they should water the field that we passed’ Note that the verbs in the subordinate clauses are marked by a special subordinative suffix, glossed here as -sub. This fact will be important in the following discussion, because it gives an explicit and unambiguous indication of whether a certain clause is independent or subordinate. Given that the origin of the Akkadian relativizer is transparently a demonstrative pronoun, Akkadian at first sight may appear to be one more example in the long line of languages that suggest the integration of a paratactic clause with a demonstrative pronoun into a relative clause structure, just as Heine and Kuteva sketch in (2) above. But as it happens, Akkadian supplies us with a few more details which make it transparent that the actual development was nothing of the sort, and that the ultimate origin of relativization was the expansion of genitival structures. The story of the Akkadian relative clauses can be sketched briefly as follows (a fuller account was given in Deutscher 2001). Akkadian has both head and dependent marking in the genitival construct. The dependent noun is marked by a genitive case ending, and the head noun is marked by the ‘construct state’ (which for convenience I gloss .of). The construct state is not shown by a particular suffix, but rather by the absence of the final -m of the case suffix on the noun, or by the absence of the whole case suffix. Thus, the noun dīn ‘judgement’ would usually appear with a case ending (nom: dīn-um, acc: dīn-am, gen: dīn-im) but in the construct state, it can be simply dīn: (7) dīn šarr-im judgment.of king-gen ‘the judgment of the king’ Now, in addition to the main productive type of relative clauses in Akkadian, which was shown above, and which I will henceforth call the ‘new style’ relative clauses, there was also another, older, relative construction in the language. The ‘old style’ relative clause had no demonstrative as a relative marker. In this construction, the onset of the relative clause was marked only by the construct state on the head noun: (8) tuppi addin-u-šum tablet.of I.gave-sub-to him ‘the tablet that I gave to him’ The old style relative clause was still in semi-productive use in the early periods, but became restricted to poetic and elevated styles in later stages. Now, what is the origin of the new style relative clauses, those with a demonstrative pronoun? As explained in Deutscher (2001:410–12), the new style RCs were modelled on the old style relative construction. Old Akkadian examples such as (9) below show that the demonstrative in the

Nominalization and the origin of subordination 

new style RC must have started out in life, very simply, as a pronominal head of an ‘old style’ relative clause. This demonstrative must have been perceived to be in the construct state, and it simply functioned as a head of the relative clause, just as any noun would: (9) šūt [in tu.ra uḫ̮irū-n] līḫuz those(acc.m.pl).of [in illness were.delayed-sub] he.should.take ‘he should take those who were delayed because of illness’ What should be clear is that the demonstrative pronoun was, from the very beginning, the head of a subordinate clause, as can be seen from the subordinative form of the verb ‘were.delayed’. The new style relative clauses must have started out in life as some sort of appositional pattern: ‘the judgement, the one which he gave [… should not be changed]’. However, this was not the juxtaposition of two independent clauses, but the apposition of an already subordinate relative clause (old style), which had a demonstrative (rather than a noun) as its head. This structure is shown in 10a. Then, through a process of integration, the demonstrative pronoun was degraded from an independent head of a relative clause to a mere marker of the onset of the relative, as in 10b: (10) a.

before integration dīn-um šu [idīn-u] judgment-nom that (one).of [he.rendered-sub] ‘the judgment, that (one)[(which) he rendered]’

b.

after integration dīn-um [šu idīn-u] judgment-nom [rel he.rendered-sub] ‘The judgment[that he rendered]’

Akkadian thus shows that the demonstrative pronoun became a marker of relativization not through a process of integration of two independent clauses, but rather through the integration of an independent clause with an already existing relative clause, one which was originally headed by this demonstrative. So the process that turned a demonstrative into a relativizer was not the genesis of relativization, but only the renewal of a marker in an already existing subordinate structure. The old marker for the onset of the relative clause was the construct state on the head noun, while the new marker was an erstwhile demonstrative. What, then, is the real origin of relativization in Akkadian? Here we have to look at the origin of the ‘old style’ relative clauses, those introduced by the construct state on the head noun. And in fact, their origin is not difficult to guess when one compares (7) and (8) above. There is an exact parallel between the grammar of genitival constructs and of relatives clauses (see further in Deutscher 2001). A similar parallel between genitives and relatives has of course been demonstrated for Tibeto-Burman by Matisoff (1972) and in a lot of subsequent literature, and has also been pointed out for many languages by Aristar (1991:9ff.).

 Guy Deutscher

The most plausible origin of the old style relative clauses in Akkadian is thus the expansion of genitival constructions. Unfortunately, we cannot follow this process of expansion historically, because the old style relative clauses are extremely old: they must go back at least to Proto-Semitic, as they are attested in the earliest strata of most Semitic languages (cf. Lipinski 2001:533ff., and Givón 1991:267–8 for early Biblical Hebrew examples). However, it is not difficult to imagine how this process of expansion could have proceeded, through the agency of nominalized verbal forms. For example, the initial stage could have been a nominalized verbal form traditionally called ‘infinitive’ (which inflects for case) that functioned as a simple nominal element of a genitive construct, as in (11a). It could then have started expanding, for instance by acquiring a pronominal argument (as in 11b). And on exactly the same lines as the developments demonstrated by Heine (this volume), the construction would then gradually acquire more verbal characteristics, until it fully expanded to a finite subordinate clause, as in (11c): (11) a

bussurat had-êm tiding.of rejoice.inf-gen ‘tiding of rejoicing’

b.

awāt had-ê-ka word.of rejoice.inf-gen-2msg.poss ‘word/matter of your rejoicing’ (a matter about which you should rejoice)

c. awāt tadbub-u word.of you.discussed-sub ‘word/matter that you discussed’ Although the actual steps in this process occurred far too early for us to reconstruct them in any accuracy, the complete parallel between the genitive and relative constructions leaves little doubt that the relative construction arose from the genitive. Akkadian thus shows that appearances can be deceptive. The productive type of relative clauses in the attested period (the ‘new style’) is introduced by what had originally been a demonstrative. But this demonstrative did not get to there through the integration of two erstwhile independent clauses. Rather, it got there through the integration of a main clause with an already existing subordinate structure, one whose ultimate origin was the expansion of a genitive construct.

2.3 Is Akkadian the exception or the rule? The main question we need to ask is whether the development in Akkadian was unusual, or whether it is representative of the many languages where relative clauses with demonstratives are claimed to have arisen through integration. Most

Nominalization and the origin of subordination 

of the evidence adduced for the integration of paratactic clauses in the ‘demonstrative channel’ has come from the Germanic languages, so it is especially useful to examine the situation there. (For expositions of relativization in the early stages of the different Germanic languages, see e.g., Stong-Jensen 1977; Mitchell 1985; Hock 1991; Harbert 1992; Pittner 1995.) While the Germanic languages differ from one another in important details and present a picture that is far from clear, the general situation in early Germanic seems to be similar to that in Akkadian. This can be illustrated by examples from Old Icelandic (Stong-Jensen 1977). In Icelandic, there was an existing relative clause structure with an invariable relative particle es, which could introduce relative clauses on its own: (12) …vóro þar þeir menn [es Norðmenn kalla Papa] were there those men [rel Northmen call Papa] ‘there were there those men that Northmen call Papas’ (Stong-Jensen 1977:14) There were also relative clauses which were based on the model of (12), but which were headed by a demonstrative pronoun: (13) ok blótaðe hrafna þriá þá and worshipped ravens three.acc.m.pl those.acc.m.pl

[es hánom skylldo leið visa] [rel him should way show]

‘and he worshipped three ravens, those that should show him the way’ (Stong-Jensen 1977:13)

The relation between (12) and (13) is parallel to the relation between the older and the newer types of relative clauses in Akkadian. (12) is equivalent to Akkadian relative clauses introduced only by the construct state on the head noun (as in 8 above), and (13) is equivalent to the newer type of relative clauses, which were based on the old model, but originally had a demonstrative as their head (as in 10a, or 10b, depending on whether one considers the Icelandic examples to have undergone integration or not). In any case, it is clear that the demonstrative þá in (13) is the head of an already existing relative clause, not an element of an independent clause. The picture in Old English seems to be similar. The demonstrative þam in (14) is the head of an already existing relative clause, one formed with an invariable particle ðe (corresponding to Icelandic es): (14) Ða wæs æt ðam geongan grim andswaru eðbegete then was for the.dat.m.sg youth.dat.m.sg grim answer easy.to.get þam [ðe ær his elne forleas] dem.dat.m.sg [rel earlier his courage lost] ‘then was for the youth a grim answer was easy to get, (for) the one that earlier lost his courage’ (Hock 1991:56)

 Guy Deutscher

Thus, the Germanic demonstrative pronouns which later came to be the sole markers for the onset of the relative clause seem to have started, just as in Akkadian, as heads of already existing relative clauses. The ultimate origin of relative clauses in Germanic may also be expansion, and thus nominalization. (In fact, Lehmann 1984:378 suggests a scenario for relativization through expansion for Old German.) The ‘demonstrative channel’ for the genesis of relative clauses through integration, allegedly exemplified by relative markers of demonstrative origin, may therefore turn out to be mostly a mirage.

2.4 Conclusion Nominalization has so far largely escaped the net cast by grammaticalization studies, yet it is a crucial element in the genesis of subordination. It is the core element in the channel of ‘expansion’, and it may also turn out to be behind the genesis of relative clauses that have been claimed to have arisen through ‘integration’. Any attempt to explain the genesis of subordination can thus only begin to make sense if it explains the origins of nominalization, and if it shows how the ability to repackage a verb as a noun arises in contexts where it had not existed before. I tentatively suggested one way that can account for the genesis of nominalization without already presupposing it, based on back-formation from the process of verbalization. But the subject requires far more attention.

References Aikhenvald, A. 2002. Classifiers: A Typology of Noun Categorization Devices. Oxford: OUP. Aristar, A.R. 1991. On diachronic sources and synchronic pattern: An investigation into the origin of linguistic universals. Language 67: 1–33. Blake, B.J. 1999. Nominal marking on verbs: Some Australian cases. Word 50(3): 299–317. Bybee, J.L., Perkins, R.D. & Pagliuca, W. 1994. The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the world. Chicago IL: University of Chicago Press. Carlson, R. 1994. A Grammar of Supyire. Berlin: Mouton de Gruyter. Clark, E.V. & Clark, H.H. 1979. When nouns surface as verbs. Language 55: 767–811. Cohen, D. 1984. La phrase nominale et lévolution du système verbal en sémitique. Études de syntaxe historique. Leuven: Peeters. Comrie, B. & Thompson, S. 2007. Lexical nominalization. In Language Typology and Syntactic Description, 2nd Edn. Vol. 3, T. Shopen (Ed.), Cambridge: CUP. DeLancey, S. 1986. Relativization as nominalization in Tibetan and Newari. Paper presented at the Nineteenth Annual International Conference on Sino-Tibetan Languages and Linguistics, Ohio State University, Columbus OH. Deutscher, G. 2000. Syntactic Change in Akkadian: The Evolution of Sentential Complementation. Oxfod: OUP. Deutscher, G. 2001. The rise and fall of a rogue relative construction. Studies in Language 25: 3405–422.

Nominalization and the origin of subordination 

Deutscher, G. 2005. The Unfolding of Language. New York: Metropolitan Books. Genetti, C. 1991. From postposition to subordinator in Newari. In Approaches to Grammaticalization, E.C. Traugott & B. Heine (Eds), 227–256. Amsterdam: John Benjamins. Givón, T. 1991. The evolution of dependent clauses in Biblical Hebrew. In Approaches to Grammaticalization, Vol. II, E.C. Traugott & B. Heine (Eds), Amsterdam: John Benjamins. Givón, T. 1994. Nominalized Clauses in Ute: The diachronic seesaw of finite and non-fine structure, 2ndo Encuentro de Linguistica en el Noroeste, Hermosillo: Universidad de Sonora. Givón, T. (in this volume). Multiple routes to clause: The diachrony of complex verb phrases. Greenberg, J. 1991. The last stages of grammatical elements: Contractive and expansive desemanticization. In Approaches to Grammaticalization, Vol. 1, E.C. Traugott & B. Heine (Eds), 301–14. Amsterdam: John Benjamins. Harbert, W. 1992. Gothic relative clauses and syntactic theory. In On Germanic Linguistics. Issues and Methods, I. Rausch, G.F. Carr & R.L. Kyes (Eds), Berlin: Mouton de Gruyter. Harris, A.C. & Campbell, L. 1995. Historical Syntax in Cross-Linguistic Perspective. Cambridge: CUP. Haspelmath, M. 1989. From purposive to infinitive - a universal path of grammaticization. Folia Linguistica Historica 10: 287–310. Heine, B. & Kuteva, T. 2002. World Lexicon of Grammaticalization. Cambridge: CUP. Heine, B. & Kuteva, T. 2007. The Genesis of Grammar: A Reconstruction. Oxford: OUP. Hock, H.H. 1991. On the origin and development of relative clauses in early Germanic. In Stæfcræft. Studies in Germanic Linguistics, E.H. Antonsen & H.H. Hock (Eds), 55–90. Amsterdam: John Benjamins. Hopper, P.J. & Thompson, S.A. 1985. The iconicity of the universal categories noun and verb. In Iconicity in Syntax, J. Haiman (Ed.), Amsterdam: John Benjamins. Hopper, P.J. & Traugott, E.C. 2003. Grammaticalization, 2nd Edn. Cambridge: CUP. Jespersen, O. 1948. Growth and Structure of the English Language, 9th Edn. New York NY: Doubleday Anchor Books. Kluge, F. 1995. Etymologisches Wörterbuch der deutschen Sprache, 23 Aulgae, bearbeitet von E. Seebold. Berlin: De Gruyter. Koptjevskaja-Tamm, M. 1993. Nominalizations. London: Routledge. Koptjevskaja-Tamm, M. 2003. Action nominal constructions in the languages of Europe. In Noun Phrase Structure in the Languages of Europe, F. Plank (Ed.), 723–759. Berlin: Mouton de Gruyter. Kouwenberg, N.J.C. Forthcoming The Akkadian Verb and its Semitic Background. Winoma Lake IN: Eisenbrauns. Lehmann, C. 1984. Der Relativsatz: Typologie seiner Strukturten, Theorie seiner Funktionen, Kompendium seiner Grammatik. Tübingen: Narr. Lipinski, E. 2001. Semitic Languages. Outline of a Comparative Grammar, 2nd Edn. Leuven: Peeters. Matisoff, J. 1972. Lahu nominalization, relativization, and genitivization. In Syntax and Semantics, Vol. I, John Kimball (Ed.), 237–57. New York NY: Seminar Press. Meyer-Lübke, W. 1966. Historische Grammatik der französischen Sprache. Zweiter Teil: Wortbildungslehre. Heidelberg: Winter. Mitchell, B. 1985. Old English Syntax. Oxford: Clarendon Press. Munske, H.H. 1964. Das Suffix *-inga/-unga in den germanischen Sprachen. Marburg: Elwet Verlag Pittner, K. 1995. The case of German relatives. The Linguistic Review 12: 197–231.

 Guy Deutscher Post, M.W. 2008. Nominalization-based constructions in Tibeto-Burman: Synchronic and diachronic perspectives. Seminar presented at MPI Nijmegen, May 16, 2008. Stong-Jensen, M. 1977. The syntax of es-relatives in Old Icelandic. Cahiers Linguistiques dOttawa 5: 1–26. Szemerényi, O. 1996. Introduction to Indo-European Linguistics, 4th Edn. Oxford: OUP. Trask, R.L. 1995. On the history of the non-finite verb forms in Basque. In Towards a History of the Basque Language, J.I. Hualde, J.A. Lakarra & R.L. Trask (Eds), 207–34. Amsterdam: John Benjamins. Woodworth, N.L. 1991. From Noun to Verb and Verb to Noun: A Cross-linguistic Study of Class-changing Morphology. Ph.D. dissertation, University of New York at Buffalo.

The co-evolution of syntactic and pragmatic complexity Diachronic and cross-linguistic aspects of pseudoclefts Christian Koops & Martin Hilpert

Rice University/Freiburg Institute for Advanced Studies This chapter examines the diachronic rise of a syntactically and pragmatically complex construction type: pseudoclefts. Given that cleft constructions combine available components of grammar — relative clauses and copular clauses — do they arise in full-fledged form? If they emerge gradually, what constrains their development? We first present a corpus-based analysis of the history of English pseudoclefts and develop qualitative and quantitative measures to identify properties of pseudoclefts at different developmental stages. We then apply the same measures of grammaticalization in a synchronic comparison of pseudoclefts in contemporary spoken and written German, Swedish, and English in order to test their cross-linguistic validity. We find that pseudoclefts develop gradually in a process driven by the pragmatic exploitation of their presuppositional structure.1

1. Introduction: Cleft constructions and grammaticalization The diachrony of information structure constructions has been among the foundational issues of functional typological linguistics as well as research on grammaticalization. The synchronic finding that morphological focus markers often resemble copulas and that the non-focused (presupposed) part of focus constructions often

1. We would like to thank the editors of this volume, Tom Givón and Matt Shibatani, as well as the other participants at the 12th Rice University Symposium on Language, March 27–29, 2008, for their valuable comments on the ideas presented in this paper. We also thank the audience at the Workshop on Emergent Constructions at the Freiburg Institute for Advanced Studies, May 8–9, 2008. All remaining errors are our own.

 Christian Koops & Martin Hilpert

exhibits properties of relative clauses allowed the reconstruction of a diachronic process in which morphological focus marking systems develop from syntactic (especially cleft) constructions (Givón 1979; Heine & Reh 1984). This cleft-to-focus marker pathway, as an example of the simplification of bi-clausal to monoclausal syntax, has been the phenomenon of primary interest thus far. Here, we take a different approach to the topic. Rather than viewing focus constructions from the perspective of the reduction of syntax to morphology, we examine the emergence of cleft constructions themselves. The reason for the small amount of attention paid to the diachrony of clefts may lie in a view of cleft constructions as, in the words of Harris and Campbell (1995: 54), “universally available syntactic constructions.” While not every language has clefts, it is likely that such constructions are easily added to grammars. The focus cleft may be so widely available because it is structurally equivalent to a copular clause with a relative clause modifying one of its constituents (56).

If clefts are structurally equivalent to their diachronic sources — a copular clause and a relative clause — it would seem that they can reveal little, if anything, about the rise of grammatical complexity, as there is no need for a development to take place. In fact, they might be expected to arise more or less instantaneously and in full-fledged form as speakers re-arrange existing grammatical resources for a new purpose. Harris and Campbell suggest that even universally available constructions like clefts undergo some development, however. They start out as “exploratory” expressions that may “catch on.” Crucially, “[o]nly when the expression is used in additional contexts and is generalized … may we speak of a grammatical change having taken place.” (54, emphasis added) It remains unclear what characterizes the exploratory phase in the case of cleft constructions. Also, what is the original context of use and what are its extensions? And what motivates the process of generalization that results in a completely productive grammatical construction? In this paper, we examine in detail one type of cleft construction, pseudoclefts, in order to address these questions. Our aim is to develop a grammaticalization scenario that allows less fully developed pseudoclefts to be distinguished from more grammaticalized ones. We first present a quantitative, text-based analysis of the 300-year history of English what-clefts in order to identify relevant properties of diachronically earlier and later instances of this construction (Section 2). Next, we apply the findings of the historical analysis to synchronic, cross-linguistic data in another series of corpus analyses. We compare the present-day English pseudoclefts with those of German and Swedish, and show that the synchronic properties of German and Swedish pseudoclefts coincide well with those seen at earlier stages of English (Section 3). Finally, we interpret our findings in terms of Lambrecht’s (1994) notions of presuppositional structure and its pragmatic accommodation (Sections 4 and 5).

Diachronic and cross-linguistic aspects of pseudoclefts 

1.1 Form and function of pseudoclefts Clefts are traditionally understood as information structure constructions used to pragmatically structure a proposition into two parts: a presupposition and a focus (Prince 1978; Lambrecht 2001; inter alia). Our working definition of a cleft construction, following Lambrecht, is a complex sentence structure consisting of a matrix clause headed by a copula and a relative or relative-like clause whose relativized argument is coindexed with the predicative argument of the copula. Taken together, the matrix and the relative express a logically simple proposition, which can also be expressed in the form of a single clause without a change in truth conditions (2001: 467).

Besides the syntactically unmarked sentence structure in (1), speakers of present-day English have two focus-initial clefts and one focus-final one at their disposal, as illustrated by Lambrecht’s examples in (2).

(1) I like champagne.

(2) a. It is champagne (that) I like. (it-cleft) b. What I like is champagne. (wh-cleft) c. champagne is what I like. (reverse wh-cleft) Our concern in this chapter is the wh-cleft construction in (2b). We follow common usage in referring to it as pseudocleft and to the initial constituent as wh-clause even though, strictly speaking, we only deal with the most common type, wh-clefts built on what. The wh-clause expresses a presupposed open proposition, i.e., a proposition with a missing argument (e.g., ‘I like x’), and the focus phrase provides a value for the variable in the presupposed open proposition (e.g., ‘champagne’). The function of the construction is to specify the content of the focus phrase as the value for the variable contained in the wh-clause (e.g., x = ‘champagne’) Their specificational function, i.e., the fact that they specify the value of a variable, distinguishes pseudoclefts like (3) from superficially similar predicational copular constructions like (4).

(3) What he bought was champagne.

(4) What he bought was expensive.

(Lambrecht 2001: 494)

The referentiality of the wh-clause distinguishes (4) from (3). As Lambrecht points out (2001: 494), whereas what he bought in (4) refers to a specific, previously known object or set of objects, the wh-clause in (3) is a (non-referential) propositional function. The point of uttering (4) is not to characterize a referent but to inform the hearer of what was bought. While the English it-cleft construction has received some attention from a historical perspective (e.g., Ball 1991, 1994), the only previous study of the diachrony of

 Christian Koops & Martin Hilpert

pseudoclefts in English that we are aware of is Traugott’s (2008) recent exploratory work on three pseudocleft types: all-clefts, what-clefts, and reverse what-clefts. Using data from a drama corpus, Traugott tests the specific hypothesis, derived from the work of Kim (1995) and Hopper (2001), that these constructions first emerged in particular interactional contexts. One of her findings is that while such a development may have occurred in the history of all-clefts, the same case cannot be made for what-clefts.

2. The diachronic development of English pseudoclefts We begin by tracing the development of pseudoclefts in English from their first appearance in texts at the end of the 17th century to their status in late 20th-century written and spoken discourse. Our data come from five corpora. The first two are the Penn-Helsinki Parsed Corpus of Early Modern English (PPCEME, Kroch et al. 2004) and the Corpus of Late Modern English Texts (CLMET, De Smet 2005). These are each composed of three subcorpora containing texts from consecutive 70-year periods of Early and Late Modern English, respectively. The second and third corpora represent mid and late 20th-century written British English: the Lancaster-Oslo/Bergen Corpus and the Freiburg-LOB corpus (LOB and FLOB, Hofland et al. 1999). They contain matched quantities of text samples from identical discourse genres, published in 1961 and 1991, respectively. Finally, we present data from the Santa Barbara Corpus of Spoken American English (SBCSAE, Du Bois et al. 2000–2005), which consists of naturally occurring conversations recorded primarily in the 1980s. To ensure that the pseudocleft tokens in each corpus were exhaustively identified, we performed a complete extraction of all instances of the word what (or whate, in the earliest periods), and then manually inspected the (thousands of) hits. This methodology not only ensured that no pseudoclefts were overlooked, but also made it possible to retrieve and quantify relevant constructional variants and other related constructions, as discussed in the following sections.

2.1 Discourse frequency Figure 1 shows the basic text frequency of pseudoclefts across seven of the historical subperiods spanned by our first four corpora. With the exception of the unexpected spike in the 18th century, Figure 1 shows a gradual, sustained frequency increase, which continues through the 20th century. Its trajectory suggests that the underlying development is still ongoing, i.e., that the use of the construction continues to expand. A basic frequency count as shown in Figure 1 may of course include unwanted genre effects. The rate of occurrence of pseudoclefts in a particular corpus may, for example, be greater simply because the contained texts are of a genre which for one reason or another favors its use. The risk of genre effects is particularly high in the

Diachronic and cross-linguistic aspects of pseudoclefts  160 140 120 100 80 60 40 20 0

1570–1640 1640–1710 1710–1780 1780–1850 1850–1920

1961

1991

Figure 1. Basic text frequency (number of occurrences per 1 million words).

earlier periods, which are represented by only a few dozen texts. We therefore performed a second, relative frequency analysis designed to neutralize the variable of genre. In this analysis, the construction’s rate of occurrence is not related to the total number of words in the corpus (as in Figure 1) but to the frequency of a related and (presumably) stable construction. We were able to take advantage of the fact that pseudoclefts, like the early example in (5), were predated by predicational copular constructions with referential wh-clause subjects, as in (6) (cf. (3) and (4), see also Traugott 2008).

(5) But what I wonder at is this: I find I did not start at his Proposal, as when it came from one whom I contemn’d. (specificational, i.e., pseudocleft)

(6) But what they do is so much above my understanding, I can’t pretend to give an account of it. (predicational) (both examples PPCEME 3)

To the extent that both constructions are subject to the same or similar genre effects, such effects should thereby be neutralized. The results of the relative frequency analysis are presented in Figure 2. It shows that at the time that pseudoclefts are first attested they were used four times less frequently than predicational copular constructions with wh-clause subjects. By the 20th century, the ratio is reversed. Again, the data suggest a continuous frequency increase, except that unlike in Figure 1 no clear difference between the mid and late 20th century is discernable. In summary, two frequency measures point to the same general picture: a sustained growth, suggesting that the development of English pseudoclefts has been progressing steadily over the past 300 years. Their rise in frequency suggests a continued expansion to new contexts of use.

 Christian Koops & Martin Hilpert pseudoclefts

predicational constructions

100% 80% 60% 40% 20% 0%

1640–1710 1710–1780 1780–1850 1850–1920

1961

1991

Figure 2. Relative frequency of pseudoclefts and predicational copular constructions with wh-clause subjects.

2.2 Early pseudocleft variants Before proceeding, we need to clarify the status of a non-canonical type of pseudocleft that we included in our count. From the time of their first appearance, pseudoclefts like example (5) above were used alongside a “copula-less” variant, shown in (7) and (8).

(7) And what rendered it the more dreary when we passed, there was a thick fog that hindered us from seeing above twenty yards from the carriage.

(8) In truth, I can find no excuse for you, and, what is more, I am certain you can find none for yourself. (both examples CLMET 1)

Examples (7) and (8) show preposed what-relatives whose relativized argument is co-referential with the following clause. Such cases differ from canonical pseudoclefts in their lack of syntactic integration, i.e., the absence of a copula and the fact that the clausal antecedent does not have the form of a syntactic argument. The only link between it and the wh-clause is the co-reference relation. Strictly speaking, these constructions fall outside of our initial definition of clefts. Our decision to analyze these cases as pseudoclefts is based on their functional near-equivalence to canonical pseudoclefts. Crucially, they share the specificational function. It should also be pointed out that non-integrated pseudoclefts, including cases of copula “omission,” are well-attested in spoken English (Weinert & Miller 1996; Hopper 2001; Koops & Ross-Hagebaum 2008). An example is given in (9).

(9) What they did, they took the stubs and they cleaned them up.

(SBCSAE)

Diachronic and cross-linguistic aspects of pseudoclefts 

The variant without a copula is even more common in spoken German (Günthner 2007, see also below). German also has both variants, and as in English the canonical variant is preferred in writing. “Copula-less” pseudoclefts like (7) and (8) were most frequent in the 18th century, when they made up almost half of all attested tokens. Since then, they have gradually disappeared from texts. The two 20th-century corpora include only the formulaic what is more and what is worse. Their brief popularity and subsequent decline accounts for the frequency bump in the 1710–1780 period in Figure 1. These facts suggest that, as the canonical, “copula-full” pseudocleft gained in currency, cases like (7) came to be perceived as non-standard by comparison. Except for highly conventionalized phrases like what is more they were henceforth avoided in writing. Perhaps the lack of an overt syntactic link between the wh-clause and the focused clause made them appear too colloquial. In this respect, their status can be compared to that of left-dislocation, which is also largely banned from written genres despite its frequent use in spoken discourse.

2.3 Modification of the wh-clause Free-standing what-relatives like those in (7) and (8) almost invariably expressed some form of comparison. Typically, the second of two conjuncts was characterized as having more or equally as much of some property as the first one. In fact, this was true for the wh-clauses of early pseudoclefts overall. They often contained an adverb of comparison. (10) But what most alarmed him was a hint that it was in her (Miss Matthews’s) power to make Amelia as miserable as herself. (CLMET 1, underlining added) Another common form of modification were adverbials expressing an exceptional degree of some property. (11) But what particularly engaged our attention was a sealed note, superscribed, ‘The copy of a letter to be sent to the two ladies at Thornhill-castle.’ (CLMET 1, underlining added) In fact, early pseudoclefts only rarely occurred without some additional modification of the wh-clause. The attested types of modifiers are summarized in Table 1. Table 1. Types of modification of the wh-clause. Comparison Exceptional degree Deixis Temporal adverbs Anaphoric so Epistemic modality Addition

What concerned her even more/equally … What is especially/particularly/very clear … What is crucial here … what he needed now … What finally convinced him … what first struck me … What made it so difficult … What I actually wanted … what may appear odd … What she could also do … what is notable too …

 Christian Koops & Martin Hilpert

The uniting feature of the various expressions in Table 1 is that they implicitly or explicitly link the wh-clause proposition, and by extension the focused constituent, to the preceding discourse. For example, the wh-clause what most alarmed him in (10) implicitly makes reference to other causes for alarm. Similarly, the adverb particularly in (11) indexes other noticeable events previously mentioned in the narrative. Speaking more generally, these modifiers show that a prominent function of the early pseudoclefts was that of establishing topical coherence. In this respect, their use did not differ very much from the way pseudoclefts are used today. In written texts, pseudoclefts often occur at topic transition points as writers turn from one idea to the next (Jones and Jones 1985). In conversation, they are often used to re-orient the course of the ongoing talk, for example when resuming a temporarily suspended topic after a digression (Kim 1995). One major point of difference between pseudoclefts then and now, however, is the degree to which the link between the prior discourse and the focused constituent was routinely made explicit, as seen by the near-obligatoriness of some additional modification used to this effect. The rate of occurrence of the modifiers in Table 1 provides a measure of the diachronic generalization of the construction. Figure 3 shows how the proportion of modified and unmodified wh-clauses in pseudoclefts shifted. Whereas in the earliest period only one out of five wh-clauses was unmodified, the rate of unmodified pseudoclefts in written discourse today has risen to about three in five cases, or 60%. (In order to base the measure on large enough numbers of examples, only the data from the 1710–1780 period onwards is included.) unmodified

modified

100% 80% 60% 40% 20% 0%

1710–1780

1780–1850

1850–1920

Figure 3. Proportion of unmodified and modified wh-clauses.

1961

1991

Diachronic and cross-linguistic aspects of pseudoclefts 

2.4 Wh-clause predicates A second characteristic feature of early English pseudoclefts was that the types of predicates typically used in the wh-clause were chosen from a narrow semantic range. Table 2 lists the most frequent predicates in the 1710–1780 period. It shows that adjectives predominated, and that those verbs which were used frequently expressed similar evaluative notions as the adjectives. In general, the wh-clause expressed an evaluation of the focused constituent on a small number of semantic scales: quantity of some stated or understood property (what is more, what will suffice), exceptionality (what was remarkable, what struck me), pleasantness (what I liked, what was worse), and importance (what matters, what is of importance). Table 2. Early wh-clause predicates (occurring two or more times). 1710–1780 be more surprise (me) be remarkable be worse like add to be a hardship be astonishing be of importance be ridiculous be surprising contribute suffice

10 4 3 3 2 2 2 2 2 2 2 2 2

The early dominance of evaluative predicates provides another indicator of the construction’s subsequent generalization. Tables 3, 4, and 5 show how the spectrum of wh-clause predicates expanded over time, and came to include predicates that are less evaluative and more eventive. Utterance verbs (say, mean, tell) soon appear in the uppermost ranks, next to perception verbs (see, hear) and cognition verbs (want, know). Finally, the semantically most general verbs do and happen (shown in small capitals) enter the top frequency range, with do becoming the second most frequent predicate in the 20th century. The verbs do and happen are of particular interest in connection with pseudoclefts because in contemporary spoken English they alone make up the lion’s share of all instances in usage (Collins 1991; Hopper 2004). Our own spoken English data also shows this strong skew towards do and happen, as can be seen in Table 6. It is interesting, then, that historically, and in terms of frequency, do and happen are relative

 Christian Koops & Martin Hilpert

Tables 3, 4 and 5. Later wh-clause predicates. 1780–1850 be worse be more like do mean want strike (me) tell

1850–1920 7 6 4 4 4 4 3 3

want be more say do mean know see be worse like propose puzzle (me)

1961, 1991 22 19 17 16 8 8 8 5 4 4 4

be more do need say mean have (got) matter see happen

20 13 11 9 8 7 7 6 6

Table 6. Most frequent wh-clause predicates in present-day spoken American English (Santa Barbara corpus). 1980s (spoken) do happen say be be funny be good get (me) remember talk about

55 17 6 5 3 2 2 2 2

latecomers.2 Moreover, there seems to have been a stepwise introduction of them into the construction, with do preceding happen, as shown in Table 7. We will return to the relevance of this sequence in Section 4. The specialization of the construction to be used with do and happen, especially in spoken discourse, seems to be the final stage in the development of a mature pseudocleft.3 A sign of even further generalization is the use in present-day spoken (but not written) English of the maximally general verb be in pseudoclefts, as in (12). The utterance was produced in reply to the question what a brand inspection is.

2. Traugott (2008) discusses early examples of pseudoclefts with do in dramatic dialogue. 3. The strong skew towards ‘do’ and ‘happen’ is also found in spoken Tagalog, where, judging by their discourse frequency, pseudoclefts are at least as advanced as in English (Naonori Nagaya, p.c.).

Diachronic and cross-linguistic aspects of pseudoclefts 

Table 7. Proportion of do and happen as wh-clause predicates in the written corpora.

do happen Other predicates Proportion do/happen

1640–1710

1710–1780

1780–1850

1850–1920

1961, 1991

0 0 3 0%

0 0 106 0%

4 0 115 3%

16 2 231 7%

13 6 247 7%

(12) What it is is your only legal bill of sale, for horse or cattle, in the State of Colorado. (SBCSAE 56) To summarize the historical analysis, wh-clefts in English developed their modern usage potential gradually. The emergence of today’s fully productive construction can be traced on the basis of several parameters, which provide measures of its degree of grammaticalization. First, the construction’s status is reflected in the degree to which the relation between the wh-clause proposition and the preceding discourse was made explicit by the use of appropriate modifying expressions. Second, the types of predicates occurring in the wh-clause show that the construction could contain an increasing number of different open propositions. Originally, these were largely restricted to evaluations of a referent or state of affairs (e.g., what surprised me…). Later, predicates appear which express an event (e.g., what happened was…).

3. Pseudoclefts in present-day English, Swedish and German Before we discuss the motivations underlying the diachronic course of events, we take a look at pseudoclefts in two other languages, German and Swedish, in a synchronic, corpus-based comparison with English. The purpose of this analysis is to determine whether the parameters we used to quantify and chart the grammaticalization of pseudoclefts in Section 2 reflect idiosyncrasies of the history of English or whether they are more generally applicable. Can they be used to assess the status of pseudoclefts in other languages? If they are indeed more generally valid, perhaps reflecting aspects of a universal grammaticalization path, less developed pseudoclefts in other languages should echo the properties seen in English at earlier historical stages. It is well known that what can be expressed in form of a cleft in one language often is not, or not as easily, rendered as a cleft in another. The non-equivalence of even structural counterparts in closely related languages has been demonstrated in contrastive translation corpus studies. For example, M. Johansson (2001, 2002) examined the frequency with which clefts occurring in English novels are translated as clefts in the novels’ Swedish editions. The rate at which they are rendered as clefts is generally low (see also S. Johansson 2001 for English, German and Norwegian). But while it is clear that not all clefts of the same general type, for example pseudoclefts, are grammatically

 Christian Koops & Martin Hilpert

equal, there has so far been no general grammaticalization-oriented attempt to work out what exactly their differences consist in and how to explain them. For example, in a recent cross-linguistic survey, Miller (2006) notes that while English has “the full range of clefts,” other languages have only “a rudimentary cleft construction” (171). Miller’s term “rudimentary” is intuitively appealing, but its implications are unclear. Some form of development is implied but not made explicit. What exactly are such clefts lacking? And does their deficiency follow a general pattern? The driving hypothesis behind our analysis is that a large amount of the crosslinguistic variability may be reduced to different degrees of grammaticalization. Many synchronic differences between the pseudoclefts of English, Swedish, and German, for example, might be the predictable effects of pseudoclefts at different points on a common developmental trajectory. Specifically, given their more rudimentary character, we predict that German and Swedish pseudocleft constructions should consistently score lower than the English pseudocleft on our quantitative measures. Also, to the extent that German and Swedish pseudoclefts themselves are at different developmental stages, we should be able to rank them relative to each other. Our choice of German and Swedish as points of comparison is of course also based on the availability of large amounts of usage data in the form of electronic corpora, which our methodology requires. We used four late 20th-century corpora, two each for Swedish and German, spoken and written discourse. The written data are taken from the Stockholm Umeå Corpus (Ejerhed et al. 2006) and the HAMBURG corpus (Hilpert 2004), which are exactly comparable to the LOB and FLOB corpora because they were constructed according to the same sampling design. The two spoken language corpora are the Göteborg University spoken language corpus (Allwood et al. 2000) and the Freiburg (aka. Grundstrukturen) corpus (Engel & Vogel 1975).4 Our operational definition of pseudoclefts had to be adjusted to the syntactic facts of German and Swedish, which are slightly different from those of English pseudoclefts. First, in both German and Swedish the presupposed part of the construction not only takes the form of a headless ‘what’-relative (Ger. was, Swe. vad), but can also be a headed relative clause (Ger. das was, Swe. det som ‘that which’). (13) German Das was ich die ganze Zeit mit Ihnen erörtert habe war im Grunde … ‘What we’ve been discussing all this time was basically … ’ (Freiburg corpus)

4. Our German and Swedish spoken corpora are admittedly not as representative of casual conversation as, for example, the Santa Barbara corpus. They contain a substantial amount of academic discourse and other more formal speech events. However, given that pseudoclefts are known to be used more frequently in formal discourse, this should have increased their frequency, thus making the German and Swedish spoken data more similar to the English data. The large differences we find are therefore all the more remarkable.

Diachronic and cross-linguistic aspects of pseudoclefts 

(14) Swedish Det som händer är att det bubblar upp syra. ‘What happens is that acid is bubbling up.’

(Göteborg corpus)

Swedish usage further includes the variants det vad (‘that what’), vad som (‘what which’) and det vad som (‘that what which’). The data clearly show that these variants are all used in a specificational function.5 Another point of difference is that German uses complex wh-words in relative clauses where the relativized argument is a prepositional object, e.g., worum es hier geht ‘what this is about’ or wovon das abhängt ‘what this depends on’. To keep our analysis parallel to the English one, we included these complex wh-words if the relative clause containing them could be translated into English using what and a preposition. Finally, in German as in Swedish the relative clause proposition may have a pronominal antecedent, leaving the relative clause, as it were, in a left-dislocated position. (15) German Was uns bewegt, was uns Sorge macht, das ist eben dies, dass … ‘What moves us, what worries us, (that) is just this, that …’ (Freiburg corpus) (16) Swedish Det som egentligen är uppgiften det är att diskutera vad orden betyder. ‘What really is the task (that) is to discuss what the words mean.’ (Göteborg corpus) As in English, we included cases in which the relative clause and the focused constituent are not linked by a copula. We found them to be more common in German. An example is given in (17). (17) Was sehr viel wichtiger ist in unserem Zusammenhang, wenn wir … ‘What’s much more important in this connection, if we …’(Freiburg corpus) Methodologically, our corpus analysis followed the same, maximally inclusive search criteria as the historical analysis. We obtained complete concordances of the relevant wh-words, as well as Swedish det som, and then manually inspected each concordance line.

3.1 Discourse frequency Figures 4 and 5 show that in terms of discourse frequency we find a consistent pattern GER < SWE < ENG on both frequency measures. Pseudoclefts are also consistently more common in spoken discourse.

5. This is not to say that constructions with such relatives may not be predicational in other cases. In fact, the headed relatives show a greater tendency to be used in predicational constructions than the headless ones.

 Christian Koops & Martin Hilpert 500 400 300 200 100 0

GER

SWE

ENG

GER

WRITTEN

SWE

ENG

SPOKEN

Figure 4. Basic discourse frequency (number of occurrences per 1 million words). pseudoclefts

predicational constructions

100% 80% 60% 40% 20% 0%

GER

SWE

ENG

WRITTEN

GER

SWE

ENG

SPOKEN

Figure 5. Proportion of pseudoclefts and predicational copular constructions.

3.2 Modification of the relative clause Additional modification of the relative clause is very common in German, as seen for instance in example (17) above. Table 8 shows the different types of modification, which by and large coincide with those found in English (cf. Table 1). Table 9 shows the modifiers found in the Swedish data, where modification is not quite as frequent.

Diachronic and cross-linguistic aspects of pseudoclefts 

Table 8. Types of modification of the relative clause in German. Was mir viel mehr auffiel … Was mich besonders gefreut hat … Was da zu sehen ist … Was wir jetzt brauchen … Was sicher nicht stimmt … Was mich so verwunderte … Was ich damit meine … Was man dann bekommt … Was ich noch sagen möchte … Was auch fehlt …

Comparison Exceptional degree Deixis Epistemic modality Anaphoric so Other anaphor Temporal adverbs Addition

Table 9. Types of modification of the relative clause in Swedish. Vad vi gör åt det mer … Det som går upp reellt … Det som sen hände … Vad som kan hända … Det som är så synd … Det som fortfarande skiljer det … Vad som syns också här …

Comparison Exceptional degree Deixis Epistemic modality Anaphoric så Temporal adverbs Addition

As can be seen in Figure 6, the rate of unmodified pseudoclefts is consistently lowest in German. In spoken discourse, we again find the pattern GER < SWE < ENG. Interestingly, there is no clear difference between the English and Swedish written data here. It appears that in Swedish writing unmodified relatives in pseudoclefts have already reached the ceiling level of about 60%, which we saw above for written English (cf. Figure 3). unmodified

modified

100% 80% 60% 40% 20% 0%

GER

SWE WRITTEN

ENG

GER

SWE SPOKEN

Figure 6. Proportion of unmodified and modified relative clauses.

ENG

 Christian Koops & Martin Hilpert

3.3 Wh-clause predicates The predicates occurring in the relative clause provide more evidence of the same general pattern. Tables 10 and 11 list the most common predicates in German and their absolute frequency in the corpora. The reason for the scarcity of the data is the low discourse frequency of pseudoclefts in German overall. Very few predicates occur more than once. Note that evaluative predicates predominate (‘be important’, ‘regret’, ‘puzzle, surprise’) but that utterance and perception verbs (‘say’, ‘mention’, ‘see’) are well established. Conspicuously absent, at least among the high frequency predicates, are the equivalents of ‘do’ and ‘happen’. These are almost completely absent in writing, and rare in spoken discourse (see also Weinert 1995; cf. Günthner 2007). In Swedish, by contrast, göra ‘do’ and hända ‘happen’ occupy the top frequency ranks in both the written and the spoken data, as shown in Tables 12 and 13. Other common predicates in Swedish are utterance verbs (‘say’, ‘mean’, and ‘think’ in the sense of expressing an opinion) and perception verbs (‘see’, ‘distinguish’).

Tables 10 and 11. Most frequent relative clause predicates in German. German written wichtig sein ‘be important’ erwähnen ‘mention’ bleiben ‘remain’

German spoken wichtig sein ‘be important’ sehen ‘see’ bedauern ‘regret’ verwundern ‘puzzle, surprise’ sagen ‘say’

2 2 2

5 5 2 2 2

Tables 12 and 13. Most frequent relative clause predicates in Swedish. Swedish written göra ‘do’ handla om ‘be concerned with’ hända ‘happen’ vara kvar ‘be left, remain’ utmärka ‘distinguish’

Swedish spoken 5 5 4 3 3

göra ‘do’ hända ‘happen’ säga ‘say’ mena ‘mean’ tycka ‘think’ se ‘see’

20 15 9 8 6 6

Figure 7 provides a quantitative comparison of the combined proportion of the verbs ‘do’ and ‘happen’ out of all attested predicates. It shows the by now familiar pattern GER < SWE < ENG, except that once more the difference between Swedish and English is not as clear as expected in the written data.

Diachronic and cross-linguistic aspects of pseudoclefts  ‘do’, ‘happen’

other

100% 80% 60% 40% 20% 0%

GER

SWE WRITTEN

ENG

GER

SWE

ENG

SPOKEN

Figure 7. Proportion of relative clause predicates ‘do’ and ‘happen’.

In summary, the quantitative and qualitative measures of grammaticalization that helped describe the diachronic data also capture relevant cross-linguistic differences, at least within the narrow range of Germanic languages considered here. The pseudoclefts of present-day German appear to be roughly at the stage of English pseudoclefts around the year 1800, while the Swedish pseudoclefts appear to be no more than one century behind the English ones. The striking resemblance of the historical and the cross-linguistic data suggests that the properties of pseudoclefts in individual languages can indeed be regarded as reflecting cut-off points on a general grammaticalization continuum. The constraints which historically shaped the emergence of pseudoclefts in English are apparently the same as the constraints that “hold back” the corresponding constructions in German and Swedish today.

4. Discussion Our first question at the outset of this chapter was whether pseudoclefts arise instantaneously or gradually. The analysis has shown that they are the result of a gradual development that follows measurable stages. This finding does not necessarily invalidate Harris and Campbell’s claim that cleft constructions are “easily added to grammars.” After all, 300 years is not a very long time span in terms of diachronic syntax. And the reason for their fairly rapid growth may well lie in the availability of the syntactic components, as discussed in the introduction. But the data show clearly that

 Christian Koops & Martin Hilpert

the pragmatic properties of a mature cleft construction do not automatically fall out in the process of combining a relative clause and a copular clause in a novel way. It takes a sustained process of generalization of the grammatical components, or rather, of their function as part of an emerging pseudocleft, for the construction to move beyond its more “exploratory” phase. In this restricted sense, pseudoclefts can therefore not be called “universally available.” More interesting than the rate of change is the particular course that the grammaticalization process takes, and what this reveals about the constraints the emerging construction has to overcome on its way to becoming fully productive. Here we need to take a step back and reconsider what exactly it is that our measures of grammaticalization capture about pseudoclefts. Note that the changes we observed affect one particular component, the wh-clause. It starts out heavily constrained and then comes to be used more and more freely. In order to understand this process of generalization, we now briefly re-visit the information structure of cleft constructions, drawing primarily on Lambrecht’s (1994, 2001) information structure framework.

4.1 Presuppositional structure and its pragmatic accommodation As noted in Section 1, it is traditionally assumed that the use of cleft constructions requires interlocutors to share certain background assumptions. In pseudoclefts, the open proposition expressed in the wh-clause is pragmatically presupposed. Lambrecht (2001: 474) defines a pragmatic presupposition as [t]he set of propositions lexico-grammatically evoked in a sentence that the speaker assumes the hearer already knows or believes or is ready to take for granted at the time the sentence is uttered.

The set of pragmatic presuppositions conventionally associated with a grammatical construction constitute its presuppositional structure. To illustrate, the constructed exchange in (18) is a case in which the context clearly matches the pseudocleft’s presuppositional structure. (18) A: Is your knee still giving you trouble? B: Actually, what I hurt is my ankle. Speaker A’s question establishes the right presupposition, viz. that speaker B was hurt.6 Contexts that match a construction’s presuppositional structure are referred to as appropriate presuppositional situations. 6. More precisely, what is presupposed in (18) is not only that speaker B was hurt (Lambrecht’s K-presupposition), but also that this knowledge has been activated in the minds of the interlocutors and that it is a predictable topic at the point of utterance (Lambrecht’s C- and T-presuppositions; see Lambrecht 2001).

Diachronic and cross-linguistic aspects of pseudoclefts 

It is not difficult to imagine contexts that do not, or not necessarily, constitute appropriate presuppositional situations for the use of pseudoclefts. One such type, which has figured prominently in the synchronic pragmatic literature on clefts, is the use of them as “discourse openers.” For example, Prince (1978) discusses the hypothetical utterances in (19), a professor’s opening words in a lecture. (19) a. *What one of my colleagues said this morning was … b. What we’re going to look at today (this term) is … (Prince 1978: 889) The lecture context rules out (19a) because the question of what a colleague may have said is not sufficiently predictable, or in Prince’s words, “cooperatively assumable” (1978: 889). Yet (19b) is appropriate. In fact, as Prince points out, it would be appropriate in this context even if the listeners were not actually thinking about the contents of the class at all at the moment the sentence is uttered. The possibility of using a cleft construction in the absence of a presupposition that it conventionally requires, as in (19b), is captured by a caveat in Lambrecht’s definition of pragmatic presupposition (cited above). It states that hearers need to be at least “ready to take for granted” the relevant proposition. Drawing on Lewis (1979), Lambrecht refers to the act of going along with a speaker and acting as though a proposition were indeed shared as the pragmatic accommodation of presuppositional structure. In (19b), the pseudocleft’s presuppositional structure can be accommodated easily. The open proposition ‘today we’re going to look at x’ is so easy to “find” in the situational context (a lecture) that listeners can be expected to supply it themselves. The option of relying on a certain amount of pragmatic accommodation means that the range of appropriate presuppositional situations for a cleft construction is not fixed but somewhat negotiable. Whether an open proposition can be presented as presupposed depends on the amount of accommodation listeners are able or willing to perform, given the nature of the proposition and the nature of the available linguistic and non-linguistic contextual clues. We believe that this inherent flexibility is the key to understanding the course of events in the history of the English pseudoclefts.

4.2 Modification of the wh-clause Consider one more set of examples from the synchronic pragmatic literature on clefts. Declerck (1988) presents the invented utterance in (20) in support of his argument against Prince (1978) that pseudoclefts may even be used where the relevant presupposition is not in any way “cooperatively assumable.” The intended context of (20) is the beginning of a speech. Later in his book, he provides another example, shown in (21), to illustrate the same point.

 Christian Koops & Martin Hilpert

(20) What I have often asked myself is how other linguists manage to keep abreast with the rapid developments in the different fields of linguistics while still finding time to go on writing articles themselves. (21) A: I hear you’ve got a job at Johnson’s. A nice place that is. I suppose you’re happy now? B: Well, I don’t know. What I’d really like to do is run a business of my own. (Declerck 1988: 213, 216, underlining added) While the situational context may be of less help in these cases than in (19b), the principle of pragmatic accommodation still accounts for them. Specifically, what helps in accommodating the open propositions in (20) and (21) are the additional modifiers placed in the wh-clauses (whose role is not discussed by Declerck). To test their import, we invite the reader to judge the felicity of (20) and (21) if often and really were omitted. Often in (20) boosts the perceived relevance of the fact that the speaker has asked herself the particular question. Listeners are more likely to accept that ‘I have asked myself x’ out of the blue if it is a question that she has often asked herself. Really in (21) allows the proposition ‘B would like to do x’ to be accommodated more easily by indexing alternative values for the variable in the wh-clause, i.e., other career choices. One candidate is available in the linguistic context, viz. B’s job at Johnson’s. Thus, really facilitates the ad hoc construction of the required presupposition by providing a link to the prior discourse. Modifying expressions like often and really are of course among the types of modification historically found in pseudoclefts (cf. Table 1). As discussed in Section 2.3, especially the early pseudoclefts often included such modifiers in the wh-clause to connect it to the ongoing discourse. We can now re-interpret the function of these modifiers as that of facilitating the pragmatic accommodation of the wh-clause proposition. Moreover, if the purpose of additionally modifying the wh-clause was indeed to facilitate the accommodation of the pseudocleft’s presuppositional structure, the decreasing rate of occurrence of these modifiers over time suggests that their effect was most needed early on. This was followed by a period in which listeners could be increasingly counted on to accommodate the presuppositional structure of pseudoclefts even in their unmodified form. We will return to this point below.

4.3 Wh-clause predicates The gradual widening of the spectrum of attested wh-clause predicates discussed in Section 2.4 demonstrates the construction’s capacity to express an increasing number of different open propositions. In the beginning, wh-clauses with evaluative predicates (e.g., what’s true…, what surprised me…) were strongly preferred. In Kim’s (1992, 1995)

Diachronic and cross-linguistic aspects of pseudoclefts 

terminology, such wh-clauses are expressions of the speaker’s epistemic and affective stance. Interpreted in terms of the notion of pragmatic accommodation, this finding suggests that such predicates, or rather the propositions expressed by the wh-clauses containing them, were easiest to accommodate pragmatically. Why should pragmatic accommodation be more readily secured in these cases? We suggest that this is because evaluations are grounded in the speaker’s subjective beliefs, attitudes, and emotional reactions. Unlike events in the world, evaluative reactions can be taken for granted in many more situations. It is probably no coincidence that Kim (1992) points precisely to an example involving an affective stance predicate in his critique of the relevance of traditional information structure categories, like presupposition and focus, for the analysis of pseudoclefts in conversation. Regarding the pseudocleft in (22), which occurred in a conversation, he points out that “the information contained in the initial WH-clause can in no way be related to some preceding context” (1992: 23). (22) The pay goes up- what I’m surprise of is the pay goes up to sump’n like two fifty an hour. While we agree that the talk preceding (22) (not fully cited here) indeed does not include any prior indication of the speaker’s surprise, we would argue that in a case like this the relevant open proposition is simply very easy to accommodate, so much so that no linguistic context is needed. The example thereby demonstrates the advantage of evaluative predicates. Another reason for maintaining that different degrees of pragmatic accommodation were involved is that the later, non-evaluative predicates came to be used in the construction in a particular historical sequence, which would otherwise remain unexplained (cf. Section 2.4). We can see this particularly clearly in the case of the verbs do, happen, and be. The diachronic order in which they become available in pseudoclefts illustrates the general observation that the wh-clause proposition is initially more contextually dependent, i.e., that it needs to echo established aspects of the discourse to a greater extent. Note that the three verbs enter the construction in an order that reflects their degree of semantic schematicity. (23) i. ‘do’ dynamic event, participants specified ii. ‘happen’ dynamic event, participants unspecified iii. ‘be’ any event or state of affairs Although do is already semantically highly general, in order of magnitude it is the most lexically rich of the group. It covers only dynamic events and requires the participants to be made explicit (e.g., what we did was…). The verb happen, by contrast, requires no mention of any participants (e.g., what happened was…). Finally, the

 Christian Koops & Martin Hilpert

verb be is maximally general and not restricted to any particular event type (e.g., what is was was (that)… ).7 In summary, the extension of the construction to additional wh-clause predicates proceeded in such a way that the range of open propositions expressed in the wh-clause came to include cases which were increasingly less easy to accommodate pragmatically. As seen with the additional modifiers placed in the wh-clause, this suggests that listeners could be increasingly relied on to routinely perform this task.

5. Conclusion: The conventionalization of pragmatic accommodation The general picture emerging from our discussion is that from the beginning of their existence an important function of pseudoclefts was to deliberately introduce presuppositions into the discourse, i.e., to create desired presuppositional situations. In using the construction in this way, speakers regularly exploited the pseudocleft’s presuppositional structure by relying on the addressee’s capacity to accommodate increasingly less clearly established, but still recoverable presuppositions. The long-term effect that such usage can have on grammatical constructions has been noted by Lambrecht, although not in connection with pseudoclefts. The pragmatic accommodation of certain presuppositional structures may to a greater or lesser extent become conventionalized and eventually grammaticalized … It can happen that the presuppositional structure of a frequently used construction is exploited so regularly that it loses some of its force, sometimes resulting in a new meaning for the construction (1994: 70).

Thus, in conclusion, we propose that the degree to which a pseudocleft’s presuppositional structure is routinely accommodated pragmatically by listeners is the key parameter along which these constructions vary historically. A more grammaticalized pseudocleft is treated by listeners, as it were, more leniently. Listeners are ready to take for granted open propositions that they would not normally accept in the case of a more rudimentary pseudocleft. This shows that the degree to which pragmatic accommodation occurs can be conventional and construction-specific.

7. The fact that English pseudoclefts are compatible with be is particularly interesting because in many of these instances the pseudocleft combines with an “inferential” cleft, i.e., a constructions of the form it’s that. Inferential clefts operate on antecedents in discourse which are typically very low in accessibility, viz. inferences (see Koops 2007). That inferential clefts nevertheless combine with pseudoclefts bears witness to the high degree of generalization of the English pseudocleft construction.

Diachronic and cross-linguistic aspects of pseudoclefts 

References Ball, C. 1991. The Historical Development of the it-cleft. Ph.D. dissertation, University of Pennsylvania. Ball, C. 1994. The origins of the informative-presupposition it-cleft. Journal of Pragmatics 22: 603–628. Collins, P. 1991. Cleft and Pseudo-cleft Constructions in English. London: Routledge. Declerck, R. 1988. Studies in Copular Sentences, Clefts, and Pseudo-clefts. Dordrecht: Foris. Givón, T. 1979. On Understanding Grammar. New York NY: Academic Press. Günthner, S. 2007. Was ihn trieb, war vor allem Wanderlust: Pseudocleft-Konstruktionen im Deutschen. In Konstruktionen in der Interaktion, S. Günthner & W. Imo (Eds), 59–90. Berlin: Mouton de Gruyter. Harris, A. & Campbell, L. 1995. Historical Syntax in Cross-linguistic Perspective. Cambridge: Cambridge University Press. Heine, B. & Reh, M. 1984. Grammaticalization and Reanalysis in African Languages. Hamburg: Buske. Hopper, P.J. 2001. Grammatical constructions and their discourse origins: Prototype or family resemblance? In Applied Cognitive Linguistics, M. Pütz, S. Niemeier & R. Dirven (Eds), 109–129. Berlin: Mouton de Gruyter. Hopper, P.J. 2004. The openness of grammatical constructions. In Papers from the 40th Meeting of the Chicago Linguistic Society. Johansson, M. 2001. Clefts in contrast: A contrastive study of it-clefts and wh-clefts in English and Swedish texts and translations. Linguistics 39: 547–582. Johansson, M. 2002. Clefts in English and Swedish. A Contrastive Study of it-clefts and wh-clefts in Original Texts and Translations. Ph.D. dissertation, Lund University. Jones, L.B. & Jones, L.K. 1985. Discourse functions of five English sentence types. Word 36: 1–21. Kim, K.-H. 1992. WH-Clefts and Left-Dislocation in English Conversation with Reference to Topicality in Korean. Ph.D. dissertation, University of California, Los Angeles. Kim, K.-H. 1995. Wh-clefts and left-dislocation in English conversation. In Word Order in Discourse, P. Downing & M. Noonan (Eds), 247–296. Amsterdam: John Benjamins. Koops, C. 2007. Constraints on inferential constructions. In Aspects of Meaning Construction, G. Radden et al. (Eds), 207–224. Amsterdam: John Benjamins. Koops, C. & Ross-Hagebaum, S. 2008. Information structure and discourse function of amalgam wh-clefts. In Proceedings of the 37th Annual Meeting of the Berkeley Linguistics Society. Lambrecht, K. 1994. Information Structure and Sentence Form. Cambridge: CUP. Lambrecht, K. 2001. A framework for the analysis of cleft constructions. Linguistics 39: 463–516. Lewis, David. 1979. Scorekeeping in a language game. In Semantics from Different Points of View, R. Bäuerle, U. Egli & A. von Stechow (Eds), 172–187. Berlin: Springer. Miller, J. 2006. Focus in the languages of Europe. In Pragmatic Organization of Discourse in the Languages of Europe, G. Bernini & M. Schwartz (Eds), 121–214. Berlin: Mouton de Gruyter. Prince, E. 1978. A comparison of wh-clefts and it-clefts in discourse. Language 54: 883–906. Traugott, E.C. 2008. All that he endeavoured to prove was …: On the emergence of grammatical constructions in dialogual and dialogic contexts. In Language in Flux: Variation, Change and Evolution, R. Kempson & R. Cooper (Eds), 143-177. London: College Publications. Weinert, R. 1995. Focusing constructions in spoken language. Clefts, Y-movement, Thematization and Deixis in English and German. Linguistische Berichte 159: 341–369. Weinert, R. & Miller, J. 1996. Cleft constructions in spoken language. Journal of Pragmatics 25: 173–206.

 Christian Koops & Martin Hilpert

Corpora used Allwood J., Björnberg, M., Grönqvist, L., Ahlsén, E. & Ottesjö, C. 2000. The Spoken Language Corpus at the Linguistics Department, Göteborg University. Forum Qualitative Social Research 1/3. De Smet, H. 2005. A corpus of Late Modern English. ICAME Journal 29: 69–82. Du Bois, J., Chafe, W., Meyer, C. & Thompson, S. 2000. Santa Barbara Corpus of Spoken American English. Part 1. Philadelphia PA: Linguistic Data Consortium. Du Bois, J., Chafe, W., Meyer, C., Thompson, S. & Martey, N. 2003. Santa Barbara Corpus of Spoken American English. Part 2. Philadelphia PA: Linguistic Data Consortium. Du Bois, J. & Englebretson, R. 2004. Santa Barbara Corpus of Spoken American English. Part 3. Philadelphia PA: Linguistic Data Consortium. Du Bois, John & Englebretson, R. 2005. Santa Barbara Corpus of Spoken American English. Part 4. Philadelphia PA: Linguistic Data Consortium. Ejerhed, E., Källgren, G. & Brodda, B. 2006. Stockholm Umeå Corpus, Version 2.0. Stockholm: University of Stockholm, Department of Linguistics. Engel, U. & Vogel, I. (Ed.). 1975. Gesprochene Sprache. Bericht der Forschungsstelle Freiburg [Forschungsberichte des Instituts für deutsche Sprache 7]. Tübingen: Narr. Hilpert, M. 2004. The HAMBURG corpus. Houston TX: Rice University. Hofland, K., Lindebjerg, A. & Thunestvedt, J. 1999. ICAME collection of English language corpora, 2nd Edn. Bergen: HIT Centre, University of Bergen. Kroch, A., Santorini, B. & Delfs, L. 2004. Penn-Helsinki Parsed Corpus of Early Modern English. http://www.ling.upenn.edu/hist-corpora/PPCEME-RELEASE-1/

Two pathways of grammatical evolution Östen Dahl Givón (1979, 2008) has suggested three stages that characterize the diachronic rise of complex constructions: Parataxis, Syntaxis and Lexis. In this paper, it is argued that rather than having three distinct stages of grammatical evolution with a linear increase of tightness, we have to postulate different kinds of integrative processes, which tend to be interwoven with each other in complex ways, both in that they tend to take place at the same time and in that they partly presuppose each other:

a. paratactic constructions → syntactic constructions b. syntactic constructions → inflectionally marked words c. syntactic constructions → morphologically complex words In particular, then, there is an intimate relationship between (a) and (b), which means that inflectional morphology not only arises together with the phenomena that Givón labels “Syntaxis” but also to a significant extent is restricted to it.

1. Introduction It is an old thought that grammatical structures evolve by getting tighter or more condensed. In our times, it has probably been formulated in the most effective way by Tom Givón in his slogan “Today’s morphology is yesterday’s syntax”.1 While I think that the idea is in general a sound one, I would like to argue here that the evolution of tight grammatical structures is not a totally linear one but rather can go along what could be called two different pathways. I got the impulse to write this paper when I read the elaboration of the general idea of condensation in Givón (2008: ch.6), where he suggests three stages2 that in his view characterize the diachronic rise of complex syntactic structures such

. The Wikipedia entry on “univerbation” quotes it as “today’s syntax is tomorrow’s Morphology”. This is a stronger claim, since it makes a prediction rather than just recording a historical fact — and it is probably too strong at least if we take it as implying that all syntax will become morphology in due time. . Actually, Givón says “general steps”, not “stages”, but this is not consistent with how he uses the word “step” later in the paper, where “step” refers to the transitions between the three elements of (1) rather than the elements themselves.

 Östen Dahl

as complex verb phrases and relative clauses, and presumably of various other grammatical phenomena: (1) a. Parataxis: two separate intonation contours. b. Syntaxis: one single intonation contour. c. Lexis: co-lexicalization into a single word. At the end of the chapter, he reformulates this in terms of “two developmental steps”: (2) i. From paratactic to syntactic complexity. ii. From syntactic to lexical/morphological complexity. An earlier version of the same ideas was presented already in (Givón 1979) where he speaks of “processes by which loose, paratactic, pragmatic discourse structures develop — over time — into tight, grammaticalized syntactic structures”, which, however, are said to erode over time “via processes of morphologization and lexicalization”. With the caveat that “the principles motivating the erosion of syntax are not necessarily identical to those that motivate its rise”, Givón argues that “we are dealing with cyclic waves that may be characterized roughly as:”

(3) discourse → syntax → morphology → morphophonemics → zero

However, in fact, the two first stages in this cycle, that is discourse and syntax, “are often coupled (i.e., occur simultaneously)”, he says, and later in the paper (94) he strengthens this to say that “in almost every case where loose, paratactic structure is condensed historically into tight, syntactic structure, the condensation involves the simultaneous rise of grammatical morphology to better code the emergent syntax”. This is an important observation, which means that the stages in the development are not as well differentiated from each other as the schema (1) may seem to imply. and the role of what Givón calls “grammatical morphology” in the development of syntax will be important in the following. As I said in the beginning, the idea that grammatical evolution takes place by condensation is an old one. In the study of grammaticalization processes, it is often suggested that grammatical forms undergo a development which can be summarized as follows (Dahl 2004):

(4) free > periphrastic > affixal > fusional

This schema has its roots in the 18th century (with the French philosopher Condillac sometimes mentioned as the first proponent (Lehmann 1982)) and was originally thought as characterizing languages as wholes rather than individual grammatical markers. The schema in is reminiscent of Givón’s schema in (1), although focuses on grammatical markers rather than constructions. (Recall that the original understanding of the term “grammaticalization” was as the process which turns lexical words into

Two pathways of grammatical evolution 

grammatical formatives (Meillet 1912)). It may be in place to compare briefly the characteristics of the evolution of constructions and grammatical markers, respectively.

2. Non-linearity and inflectional systems For both constructions and grammatical markers, the changes that they undergo over time affect grammatical complexity in various ways. Condensation, the process which I have already touched upon in the preceding, involves moving a certain amount of structure from a higher hierarchical level to a lower one, which usually means that the structure in question has to be squeezed into a tighter spot, so to speak: two phonological phrases are replaced by one, or a phrase is squeezed into one word ((Dahl 2004) 178–180). This increases the complexity of the lower structural levels. The process of condensation can be applied to a whole construction, but may also affect just one grammatical marker and the word next to it, as when a negation morpheme cliticizes to a verb or an auxiliary. But the evolution of constructions and grammatical markers also involves other complexity-affecting changes, which can be subsumed under the rubric “growth of non-linearity” (Dahl 2004). These changes, which are particularly important in the evolution of grammatical marking, are partly covered by the last step in (4), the one from “affixal” to “fusional”. Non-linearity can be defined as everything that cannot be described in terms of the concatenation of mutually independent units (“the rosary ideal”, or if you like, the Item-and-Arrangement model). For instance, inflectional morphology can be non-linear because 1. the choice of an affix depends on the identity of the stem (as when there are several declensional classes of nouns); 2. one surface unit represents several underlying units (portmanteau morphs); 3. the borders between units are blurred (fusion); 4. stems undergo unpredictable changes from one form to another or are even wholly replaced (suppletion); 5. markings are wholly or partially prosodic, changing stress or tonal patterns and thus affecting the whole word rather than just one segment. These developments are correlated with condensation, since they are more likely to take place in tight units. They are not reducible to it, however. The rise of non-linearity fairly clearly involves an increase in the complexity of the grammatical system although it does not necessarily involve the addition of extra structure — the number of surface units may even be reduced. There are quite dramatic differences between human languages as to the amount of non-linearity in their grammars, in particular with respect to the size and character of

 Östen Dahl

the inflectional component. Also crucially, this variation appears to be correlated with the socio-history or “ecology” of the language, in that non-linearity tends to be reduced in high-contact languages, in particular creoles. Thus, in the recent debate on whether “creole grammars are the simplest in the world” (McWhorter 2001), the complexity under discussion has largely been of the kind that can be subsumed under non-linearity. Non-linearity is often seen as dysfunctional and as “historical junk”, i.e., the accidental results of “blind” diachronic processes (“erosion”). Arguably, however, nonlinearity may have processual advantages, although I won’t go into that question here. But I want to make a point that is discussed in more detail in (Dahl 2004): the rise of inflectional morphology is not adequately characterized as a result of “erosion”, for several reasons. The first one is that erosion is a bad metaphor for phonological changes that come about when an element is squeezed into a tighter slot or given a less prominent position in the structure, as this change is adaptive rather than the result of random entropy-increasing processes — I have suggested that “trimming” is a better word for an element being reduced to make it fit better into the space allotted for it. The second reason is that the rise of inflectional morphology also can involve the creation of new abstract structure. Thus, previously unrelated forms can by various processes come to be understood as belonging to the same paradigm, most clearly perhaps in the case of suppletion (e.g., English go and went), or the absence of a certain grammatical marker can come to be understood as meaningful, if that grammatical marker is used obligatorily or preferentially given that a certain condition is fulfilled (cf. (Bybee, Perkins, & Pagliuca 1994)). In addition to exhibiting non-linearity, inflectional systems also tend to have a number of quite specific properties that set them off from other parts of the language system. I argue in (Dahl 2004; Chapter 9), that these can be seen as consonant with a description according to the “Word-and-Paradigm” model, in which word forms are seen as the manifestations of a lexeme and an unordered set of morphological properties. Thus, inflectional systems tend to involve closed sets of possible forms, arrangeable in paradigms, formally n-dimensional matrices (where n is a small finite number corresponding to the number of inflectional categories). Among other things, this excludes recursivity (Matthews 1991) and multiple meaning-bearing appearances of the same morpheme; moreover, there is usually no way of changing the meaning by reordering the inflectional morphemes.

3. The intimate connection between inflectional morphology and syntax Although inflectional systems are set off from the rest of the language system by their properties, as I just said, they interact quite intimately with other components, notably the syntax and the lexicon. Grammatical gender is a paragon example. In gender systems like the ones well-known from many European languages, gender is a lexical

Two pathways of grammatical evolution 

feature of nouns and an inflectional feature of adjectives and pronouns, which agree with nouns in gender given that they have certain syntactic relations to them. In fact, if we take a definition like that of (Hockett 1958), “Genders are classes of nouns reflected in the behavior of associated words”, gender does presuppose syntax, although its direct manifestation is in morphology. Given this intimate relationship between inflectional morphology and syntax, it is no wonder that the genesis of inflection typically takes part within the development of syntactic constructions, as Givón argued in his 1979 paper. Obviously, however, even if inflectional morphology often arises at or immediately after the transition from “parataxis” to “syntaxis”, new inflections can also develop within stable syntactic constructions, where there is no evidence of an earlier “paratactic” stage. For instance, it is well known that definite articles commonly develop out of demonstrative pronouns, but even if there are languages in which combinations of demonstratives and nouns are looser than in, say, English, and it is possible that constructions of the looser type could serve as a diachronic source for tighter ones, there is to my knowledge no evidence of such developments in the languages where definite articles have developed and been morphologized. Thus, if we assume Givón’s three-stage model, we see that inflectional morphology arises at Stage 2, “Syntaxis”, together with syntactic constructions, with which it is intimately connected. The question now arises, what happens to inflections in Stage 3, “Lexis”? A priori, anything could happen: inflectional complexity may continue to grow, it may stay the same, or it may be reduced or go to zero. If inflectional complexity were a simple function of the tightness of a construction, we would expect the first to be the case. However, in fact, it seems that the growth of inflectional complexity and the development of tighter, “co-lexicalized” constructions in fact have a rather strong negative correlation, and that there is in fact evidence for speaking of two separate pathways of evolution, rather than two successive stages in one linear development. I shall use the phenomenon of differential object marking to illustrate what I have said. Cf. the following example from Southern Ute (Givón 1995): (5) a. kwana-ci ‘uway paqa-pųga eagle-AN/obj def/obj kill-rem ‘He killed the eagle’ b. kwana-paqa-pųga eagle-kill-rem ‘He did some eagle-killing’ or ‘He killed eagles’ In (5a), there is an object noun phrase kwana-ci ‘uway, which contains both a casemarking suffix -ci and a determiner ‘uway.3 In (5b), there is no independent object . Strictly speaking, ‘uway is a “remote-invisible” demonstrative which here functions as a definite article (Givón & Southern Ute Tribe 1980).

 Östen Dahl

noun phrase, rather the stem kwana ‘eagle’ shows up as an incorporated part of the verb. With respect to the way the direct object is realized, (5a) could be said to be a typical representative of “Syntaxis”, whereas (5b) exemplifies “Lexis”, more specifically the well-known phenomenon of noun incorporation. What we can note is that the object in (5b) is not connected with any type of grammatical marking. In fact, the Southern Ute sentences represent a very general pattern, in which direct objects are differentiated in such a way that high-referentiality objects get full grammatical marking while low-referentiality objects get reduced or no marking, with variation in how the two groups of NPs are delimited. Thus, in Turkish, accusative case-marking can be omitted with indefinite direct objects: (6) a. Ayşe balıği tutuyor. A. fish.acc catch.prs.3sg ‘Ayşe is catching the fish.’ b. Ayşe balık tutuyor. A. fish catch.prs.3sg ‘Ayşe is catching fish.’

(Nilsson 1985)

Such zero-marked noun phrases are restricted to the position immediately before the verb (which is sentence-final in Turkish), a fact that could be interpreted as indicating that (6b) represents a tighter construction than (6a), even if it has not reached the stage of full incorporation. These facts are of course well-known, and I could cite many similar examples. What I want to focus on here, though, is the fact that the tighter constructions (5b) and (6b) contain no grammatical markings pertaining to the direct object. In other words, at least for these cases, it looks as if whereas the step from “Parataxis” to “Syntaxis” is connected with the rise of inflectional morphology, the further step from “Syntaxis” to “Lexis” shows the opposite tendency: inflectional morphology disappears. This pattern is not restricted to direct object marking but appears to be quite general. For instance, in some Scandinavian vernaculars, attributive adjectives are frequently incorporated (this is obligatory if the noun phrase is definite), and then do not display the agreement markers found in the syntactic construction, as in the following Elfdalian examples: (7) a. gambler kaller old-pl.m.nom man-pl.nom ‘old men’ b. gamt-kaller old-man.pl.nom ‘old men’ We may then suggest that inflectional morphology is essentially a phenomenon of the “Syntaxis” stage, and thus even more intimately connected to syntax. The question now

Two pathways of grammatical evolution 

is what kind of diachrony is behind this. Again, there are alternatives: either the processes that take structures from the “Syntaxis” to the “Lexis” stage involve reduction of grammatical markings, or, structures that are thus condensated have never contained grammatical markings in the first place. If we return to direct object marking, it seems to me that the second alternative is the most likely one in most cases. In many languages with differential object marking, the source of the grammatical morpheme that marks high-referentiality objects is fairly transparent, for instance, Spanish a, which has the original meaning ‘to’ and has expanded first to be a marker of indirect objects and then to animate direct objects. Inanimate direct objects, on the other hand, which represent the low-referentiality type in Spanish, have been unmarked since the breakdown of the Latin case system. A process that fused the latter with verbs to create structures analogous to that in (5b) would not have to involve any reduction of grammatical marking. It can be argued that the factors that disfavour grammatical marking of direct objects are the same that favour a tightening of the link between direct object and verb. In general, it seems that high-referentiality noun phrases are resistant to incorporation. However, it seems that at least some reduction of inflections does take place in the transition from “Syntaxis” to “Lexis”. It is not uncommon for incorporated stems to be reduced ((7b) is a case in point), so it is no wonder that inflectional elements can also be affected by the same processes. (Croft and Deligianni ms.) argue that constructions with preposed adjectives are in many languages “tighter” than those with postposed adjectives in the same languages, and in some Romance languages, reduced forms of preposed adjectives are found. Thus, in the Italian expression il bel paese ‘the beautiful country’, the absence of the usual masculine singular adjective ending -o in bel ‘beautiful country’ can hardly be explained in any other way than as the result of phonetic reduction. There are a number of further problems here. It does appear to be if not universal so at least normal for incorporation to apply only to a subset of all direct objects, and the properties of that subset are similar from language to language. Thus, only some transitive VPs are condensated as to make their way into the tighter “Lexis” stage. But if one starts considering why this is the case, it becomes fairly obvious that the direct objects that are candidates for being incorporated must be in some sense or other more “tightly” connected to their verbs than those that are not, and that must be the case even in languages where incorporation has not taken place. That is, there must be some difference between the verb phrases kill the eagle and killing eagles in a language like English that makes it possible to explain why only counterparts of the latter are plausible candidates for incorporation. So maybe expressions in languages have some kind of inherent “tightness”, or inversely, their components have an inherent degree of independence from each other. This means that when a construction comes to be expressed in a tighter way, it may either be an adaptation to the inherent tightness it

 Östen Dahl

has always had, or it may be a response to a change in the function of the construction with an ensuing change in its inherent tightness. Thus, if we take a standard case of the transition from “Parataxis” to “Syntaxis” — the development from topic-comment to subject-predicate constructions, it is fairly obvious that there is a difference between these two types in their inherent degree of condensation, which relates to differences in the discourse role of topics and subjects. So what happens when a topic-comment construction is condensated to a subjectpredicate construction is that the former extends its domain of use to cases with a higher degree of inherent tightness (this is an example of what I call “pattern spread” in (Dahl 2004)) and undergoes changes that are conditioned by this increase (which is what I call “pattern adaptation”). It is less clear that such a description applies to the development of object noun incorporation, since in those cases, it is hard to see that there is any change in inherent tightness.

4. Two pathways The differentiation of transitive verb phrases in a language such as Southern Ute into one “syntactic” and one “lexical” construction, where only the first one involves overt object marking, suggests that thinking of “Syntaxis” and “Lexis” as two consecutive stages is at least partly misleading. In the case of direct object marking, we could equally well speak of two alternative pathways. Given a construction that combines two lexical elements A and B, grammatical development can lead to results of two different kinds: either A and B coalesce into one word, or they remain separate but grammatical markers develop that eventually may fuse with either A or B. A

A

B

B

A

A

m

B

m

A

B

m

B

Thus, in the first case, the construction unequivocally moves to the “Lexis” stage, in the second, it remains at “Syntaxis”. On the other hand, both cases involve the development of morphological complexity. So if we look at Givón’s original developmental scale in , what we have to say is that the first is wholly at the morphology stage while the second is both syntax and morphology — and of course Givón notes

Two pathways of grammatical evolution 

that the first two steps of his schema can take place simultaneously. But here is a further complication. If syntax and morphology arise at the same time, where does then that morphology come from? Does that mean that there is an immediate jump from the “discourse” or “Parataxis” stage to the morphological stage? Well, if we look closer at things, we can see that this is not in fact the case. Suppose, for instance, we have a development of the following kind, which would give rise to object marking on the verb:

Parataxis Syntaxis I know him, John Æ I-know-him John ‘I know John’ It can be seen here is that it is a simplification to say that the construction to the left of the arrow is at the “Parataxis” stage — it is really only at the top level we have a relationship that can be called paratactic, since the (so-called) right-dislocation construction joins a dislocated noun phrase with a regular transitive sentence, which must be said to be at the “Syntaxis” stage — and this is in fact crucial to the further development, given that the object pronoun in the dislocated construction is the source of the affixed object marker to the right. In other words, rather than saying that the two steps take place simultaneously, we should say that a construction and its components may have reached different degrees of condensation, and that it is the combination of these degrees that conditions the following step in the development, which involves on the one hand a step from “Parataxis” to “Syntaxis” at the level of the whole construction, and a step from “Syntaxis” to inflectional morphology with regard to the relationship between the object pronoun and the verb. Similarly, noun incorporation occurs in progressive constructions in some West Germanic languages, like the following example from colloquial German: (8) Ich bin am Eis-essen. I be.prs.3sg at_def.dat.m.sg ice-cream_eating ‘I am eating ice-cream.’ Here, the construction as a whole is still periphrastic, and thus at a syntactic stage, but the object-verb relationship is encoded at the “Lexis” level.

5. Conclusions What we have seen here is thus that rather than having three distinct stages of grammatical evolution with a linear increase of tightness, we have to postulate different kinds of integrative processes, which tend to be interwoven with each other in complex

 Östen Dahl

ways, both in that they tend to take place at the same time and in that they partly presuppose each other: a. paratactic constructions Æ syntactic constructions b. syntactic constructions Æ inflectionally marked words c. syntactic constructions Æ morphologically complex words In particular, then, there is an intimate relationship between (a) and (b), which means that inflectional morphology not only arises together with the phenomena that Givón labels “Syntaxis” but also to a significant extent is restricted to it.

References Bybee, J.L., Perkins, R.D & Pagliuca, W. 1994. The Evolution of Grammar. Tense, Aspect, and Modality in the Languages of the World. Chicago IL: University of Chicago Press. Croft, W. & Deligianni, E. Asymmetries in NP word order. Ms. Dahl, Ö. 2004. The Growth and Maintenance of Linguistic Complexity [Studies in Language Companion Series 71]. Amsterdam: John Benjamins. Givón, T. 1979. From discourse to syntax: Grammar as a processing strategy. In Discourse and Syntax, T. Givón (Ed.), New York NY: Academic Press. Givón, T. 1995. Functionalism and Grammar. Amsterdam: John Benjamins. Givón, T. 2008. The Genesis of Syntactic Complexity. Amsterdam: John Benjamins. Givón, T. & Southern Ute Tribe. 1980. Ute Reference Grammar, 1st Edn. Ignacio CO: The Ute Press. Hockett, C.F. 1958. A Course in Modern Linguistics. New York NY: Macmillan. Lehmann, C. 1982. Thoughts on Grammaticalization: A Programmatic Sketch [Arbeiten des Kölner Universalien-Projekts 48]. Cologne: University of Cologne. Matthews, P.H. 1991. Morphology, 2nd Edn. Cambridge: CUP. McWhorter, J.H. 2001. The world’s simplest grammars are creole grammars. Linguistic Typology 5(2/3): 125–166. Meillet, A. 1912. L’ évolution des formes grammaticales. Scientia (Rivista di Scienza) 12(26): 6. Nilsson, B. 1985. Case Marking Semantics in Turkish. Stockholm: Dept. of Linguistics, Stockholm University.

part ii

Child language

On the role of frequency and similarity in the acquisition of subject and non-subject relative clauses Holger Diessel

University of Jena Frequency and similarity are important determinants for the acquisition of children’s early item-based constructions. This paper argues that frequency and similarity are equally important for the development of more complex and intricate grammatical phenomena such as relative clauses. Specifically, the paper shows that the acquisition of relative clauses is crucially determined by the similarity between particular types of relative clauses and simple SVO constructions. Two specific hypotheses are proposed: First, since subject relatives have the same word order as ordinary SVO clauses, they usually cause fewer difficulties in comprehension studies than non-subject relatives. Second, while non-subject relatives are structurally distinct from SVO clauses, semantically they are expressed by prototypical transitive constructions, which arguably helps the child to learn this type of relative clause.

1. Introduction In the usage-based approach to language acquisition grammatical development is crucially determined by the child’s experience with language. Specifically, frequency of occurrence plays an important role in this approach. Every time a child encounters a linguistic expression in the ambient language it leaves a trace in memory reinforcing its mental representation, i.e., the level of entrenchment, which in turn facilitates the activation of the expression in future language use (cf. Tomasello 2003; Diessel 2007a). Another learning mechanism that plays an important role in the usage-based approach is analogy, which has been characterized as the core of human cognition (Hofstadter 2001). Analogy involves a mapping of information from one particular entity, the source, to another particular entity, the target. The mapping is based on the recognition of similarity, which can be divided into two basic types: substantial (or object) similarity, which involves the recognition of shared attributes, and structural similarity, which involves the recognition of shared structures or relationships (cf. Gentner 1983; Gentner & Median 1998).

 Holger Diessel

There is abundant evidence that frequency and (structural) similarity play important roles in the early stages of grammatical development. For instance, Bybee and Slobin (1982) showed that the acquisition of the English past tense is determined by the child’s experience with individual verbs and the phonetic similarity between the various past tense forms, and Tomasello and colleagues demonstrated that the development of children’s early item-based constructions is crucially affected by the frequency and similarity of verb-argument constructions in the ambient language (cf. Tomasello 2000, 2003; Lieven et al. 2003). However, grammar does not only consist of words and simple item-based constructions, it also includes more complex and abstract grammatical phenomena such as relative clauses. It is the purpose of this paper to show that frequency and similarity are not only important determinants of the early stages of grammatical development but are equally important for the acquisition of complex grammatical structures such as relative clauses. Specifically, the paper argues that the acquisition of subject and non-subject relatives is determined by the frequency of the various relative clauses in the ambient language and their similarity to simple sentences. The paper is divided into two parts. The first part considers the results of an experimental study that I conducted together with Michael Tomasello (cf. Diessel & Tomasello 2005). The study shows that children’s comprehension and production of relative clauses is influenced by the structural overlap between the various types of relative clauses and simple sentences. Specifically, the study suggests that subject relatives cause fewer difficulties than non-subject relatives because they involve the same constituent order as ordinary (in)transitive clauses. The second part reports the results of a new corpus study that examines the meaning and use of subject and non-subject relatives in spontaneous child language. The study shows that while non-subject relatives are structurally distinct from simple sentences, semantically they are expressed by prototypical transitive clauses, whereas subject relatives comprise a variety of constructions that do not always match the corresponding properties of simple (in)transitive clauses. The paper argues that the structural and semantic overlap of children’s subject and non-subject relatives with simple sentences helps the child to bootstrap into the intricate grammar of relative clauses. In this view, grammatical development is an incremental process whereby children acquire new grammatical constructions based on structures they already know. Before I turn to these studies, I will briefly consider the external properties of children’s spontaneous relative clauses providing the background for the subsequent analyses.

2. The external properties of children’s spontaneous relative clauses In the experimental literature on the acquisition of relative clauses, children are commonly confronted with complex sentences in which the relative clause modifies the

Role of frequency and similarity 

subject or object of a transitive main clause including a prototypical agent and patient and a verb denoting a physical activity as in examples (1) and (2) (adopted from Tavakolian 1977).

(1) The pig jumped over the horse [that bumped into the lion].

(2) The horse [that kicked the cow] pushed the donkey.

The main clauses of children’s early relative clauses are very different. As shown in Diessel (2004) and Diessel and Tomasello (2000), the development of relative clauses originates from particular constructions that are much less complex than the relativeclause constructions that have been used in most experiements. Investigating corpus data from four English-speaking children between the ages 2;0 to 5;0, they found that most of the children’s early relative clauses are attached to the predicate nominal of a copular clause (cf. example 3) or, less frequently, to an isolated noun phrase (cf. example 4).

(3) Those are bugs [dat I throw].

(4) This thing [you were watching].

Adam 3;7 Abe 3;6

In both types of constructions the relative clause functions to elaborate a referent in a pragmatically marked position. Relative clauses that are attached to the predicate nominal of a copular clause modify a referent that is presented in focus position of the copular clause (cf. examples 5–6), and relative clauses that are attached to an isolate noun phrase are commonly used to characterize a referent that is presented in response to a content question (cf. examples 7–8). Both types of constructions emerge in the communicative interaction between parent and child. They are associated with particular discourse patterns that constrain their early structure and use (cf. Diessel 2004: chap 6; see also Givón 2009). (5) *CHI: Mommy … I want to eat now and then go to the park, okay? Abe 3;5 *CHI: Daddy … this is part of the egg [I ate for breakfast]. (6) *CHI: Hey … Paul … come and look at dis machine. *CHI: Dis is a new machine [dat Paul likes]. (7) *FAT: No what did you eat? *CHI: Some apples [that were sweet]. (8) *MOT: What is that? *CHI: Place [where you put your things when you eat dem].

Adam 4;3 Abe 3;6 Adam 3;9

What is more, Diessel and Tomasello argue that although relative clauses are embedded clauses, initially they appear in constructions that are not really bi-clausal. Relative clauses that are attached to an isolate noun phrase do not occur with a complete main clause, and relative clauses that are attached to the predicate nominal of a copular clause occur with a main clause that is “propositionally empty” (Lambrecht 1988): Although

 Holger Diessel

the copular clause includes a verb it does not denote an independent state of affairs but functions to present a referent in focus position such that it can be elaborated by the relative clause. In other words, the development of relative clauses originates from constructions that contain only a single proposition. Starting from such simple sentences, children gradually acquire more complex relative-clause constructions in which main and relative clauses express two separate states of affairs (cf. Diessel 2004; Diessel & Tomasello 2000, 2005). Interestingly, similar developmental pathways have been observed in several other languages. For instance, Dasinger and Toupin (1994) noticed the predominance of presentational relative constructions in the speech of Spanish- and Hebrew-speaking children, which they collected in a picture book task, and Hudelot (1980) reports that the vast majority of children’s early relative clauses in French are attached to the predicate nominal of a copular clause. Moreover, Hermon (2004) argued that there are striking parallels in the development of relative clauses in English and Indonesian: Like English-speaking children, Indonesian-speaking children begin to produce relative clauses in structures that denote only a single state of affairs. Finally, Brandt, Diessel, and Tomasello (2008) investigated a large corpus of relative clauses in the speech of a German-speaking boy who began to use relative clauses in topicalization constructions consisting of the relative clause and an isolated head noun.1 In what follows I report the results of two studies that are concerned with the internal properties of children’s relative clauses. The first study is an experimental study that was designed to examine the effect of the relativized role on the processing and acquisition of relative clauses (cf. Diessel & Tomasello 2005); the second study is an observational study that is concerned with the meaning and use of different structural types of relative clauses in spontaneous child language.

1. Ozeki and Shirai (2005) showed that children’s early relative clauses in Japanese occur in different types of constructions; they are more often attached to the main clause subject and main clause object than children’s relative clauses in English. Interestingly, Ozeki and Shirai note that early relative clauses in Japanese are only little different from adjectives: They usually include a stative verb and involve the same morphology as adjectives (cf. Kim 1989, who found similar types of relative-clause constructions in the speech of Korean-speaking children). Since adjectives express properties rather than full propositions, Diessel (2007b) suggests that children’s early relative clauses in Japanese (and other East Asian languages) are not really bi-clausal. Like English-speaking children, Japanese-speaking children begin to use relative clauses in particular constructions that denote only a single state of affairs. In English, relative clauses originate from focus and topicalization constructions in which the main clause is propositionally empty, and in Japanese, relative clauses originate from attributive constructions in which the relative clause specifies a semantic feature of the head noun. In both types of languages children begin to produce relative clauses in constructions that contain only a single proposition.

Role of frequency and similarity 

3. Study 1 The experimental literature on the acquisition of relative clauses has concentrated on two structural types: Subject relatives, i.e., relative clauses in which the subject is relativized, and object relatives, i.e., relative clauses in which the direct object is relativized (e.g., Sheldon 1974; Tavakolian 1977; Hamburger & Crain 1982; Corrêa 1995; Kidd & Bavin 2002). However, subject and object are not the only syntactic roles that can be relativized. As can be seen in (9) to (13), the relativized syntactic role can be the subject, the direct or indirect object, an adverbial, or a genitive attribute.

(9) The man [who met the woman].

subj-relative

(10) The woman [who the man met].

obj-relative

(11) The boy [who the girl gave the ball to].

io-relative

(12) The girl [who the boy played with].

adv-relative

(13) The man [whose cat caught a mouse].

gen-relative

In subject relatives, the relativized role is indicated by a relative pronoun (or particle) that immediately precedes the verb of the relative clause; but in object and adverbial relatives the relative pronoun/particle can be omitted and the relativized syntactic role is expressed by a ‘gap’ in the argument structure. For instance, the relative clause in (14) includes a gap after the preposition at the end of the sentence. (14) The thing [I was looking for __]. In the psycholinguistic literature, it is commonly assumed that the processing difficulty of relative clauses is determined by the distance between filler and gap (i.e., the head of the relative clause and the position of the relativized element). Wanner and Maratsos (1978) argued that the human processor has to keep the information of the head in working memory until it encounters the gap, providing the information necessary to integrate the filler into the relative clause. The longer the distance between filler and gap, the harder the relative clause is to parse. Thus, on this account, subject relatives are easier to process (and easier to learn) than object relatives, which in turn are easier than adverbial relatives (cf. Clancy, Lee, & Zoh 1986; de Villiers et al. 1979; Gibson 1998; Hawkins 1987, 1994: 37–46; Keenan & Hawkins 1987; O’Grady 1997: chap 9). Note however that this account relies on the particular properties of relative clauses in English, in which the relativized role is indicated by a gap. While this strategy is also found in many other languages, it is by no means the only way to form relative clauses (cf. Givón 1990: chap 15; Andrews 2007). Across languages, relative clauses are also commonly formed by means of a case-marked relative pronoun that

 Holger Diessel

indicates the relativized role right at the beginning of the relative clause, as in the following examples from German. (15) Der Mann, der mich gesehen hat.

subj-relative

(16) Der Mann, den ich gesehen habe.

obj-relative

(17) Der Mann, dem ich das Buch gegeben habe.

io-relative

(18) Der Mann, zu dem ich gegangen bin.

adv-relative

(19) Der Mann, dessen Hund mich gebissen hat.

gen-relative

Since the relative pronoun is the first word in the relative clause, the processor does not have to keep unintegrated information in working memory while processing the relative clause. As soon as the processor encounters the relative pronoun, the filler can be integrated into the relative clause, suggesting that the processing of German relative clauses involves a different strategy than the processing of English relative clauses, which may affect the acquisition process.

3.1 Methods In order to test this hypothesis Michael Tomasello and I conducted a sentence repetition experiments in which 4-to-5 year old English- and German-speaking children had to repeat complex sentences including one of six different structural types of relative clauses: 1. Intransitive subject relatives, i.e., S-relatives (e.g., the man who slept on the couch). 2. Transitive subject relatives, i.e., A-relatives (e.g., the man who saw the woman). 3. Direct object relatives, i.e., P-relatives (e.g., the man who the woman met). 4. Indirect object relatives, i.e., io-relatives (e.g., the man who John gave the book to). 5. Adverbial relatives, i.e., adv-relatives (e.g., the man who the girl went to). 6. Genitive relatives, i.e., gen-relatives (e.g., the man whose dog chased the cat). We distinguished between transitive and intransitive subject relatives because previous studies had shown that transitivity can affect children’s comprehension of relative clauses (cf. Hamburger & Crain 1982; Goodluck & Tavakolian 1982). The test items were all of the same length and tightly controlled for various semantic, pragmatic, and structural factors. After a warm-up phase, children had to repeat a total of 41 sentences including 24 relative clauses and 17 filler items. The sentences referred to a play situation in front of the child providing a pragmatically adequate context for the use of a (restrictive) relative clause (cf. Hamburger & Crain 1982).

Role of frequency and similarity 

3.2 Results Figure 1 shows the mean proportions of the children’s correct responses to the various types of relative clauses. English

4,0

German

Correct

3,0 2,0 1,0 0,0 S

A

P

IO

ADV GEN

Pairwise comparisons A/S vs. P p < 0.001 P vs. IO p > 0.173 P vs. ADV p > 0.169

S

A

P

IO

ADV GEN

Pairwise comparisons A/S vs. P p < 0.001 P vs. IO p > 0.061 P vs. ADV p < 0.001

Figure 1. Proportion of correct responses (Diessel & Tomasello 2005).

As can be seen, in both studies subject relatives (i.e., A/S-relatives) caused fewer errors than direct object relatives (i.e., P-relatives), which in turn caused fewer errors than indirect object relatives (io-relatives) and adverbial relatives (adv-relatives); genitive relatives (gen-relatives) were almost always incorrect. The overall results are similar in the English and German study; however, they differ in the domain of object and adverbial relatives. The English-speaking children basically produced the same number of errors in response to these relative clauses, i.e., the differences were not significant (cf. Figure 1); but the German-speaking children had significantly fewer problems with direct object relatives than with indirect object relatives and adverbial relatives. In particular, the adverbial relatives caused many more problems in the German study than in the English study (The difference between direct and indirect object relatives is only marginally significant in the German study; cf. Figure 1). In order to understand the children’s difficulties with the various types of relative clauses, Diessel and Tomasello investigated the children’s errors. Interestingly, they found one very common type of mistake: When children were given an object or adverbial

 Holger Diessel

relative, they commonly converted it to a subject relative. The English-speaking children converted the relative clause by changing the word order (cf. example 20), and the German-speaking children converted it by changing the case role of the relative pronoun (cf. example 21).2 (20) TEST ITEM: This is the girl [who the boy teased at school this morning]. CHILD: This is the girl [that teased … the boy … at school this morning]. (21) TEST ITEM: Da ist der Mann, [den das Mädchen im Stall gesehen hat]. CHILD: Da ist der Mann, [der das Mädchen im Stall gesehen hat]. However, the children were not consistent in making this type of error. Sometimes they converted a given relative clause, and sometimes they repeated the clause correctly. What is more, the children often noticed that they had made a mistake and repaired the conversion error before the end of the sentence (cf. examples 22–23), indicating that at least some of the children were able to produce object and adverbial relative clauses correctly despite the fact that they often changed them to subject relatives. (22) This is the girl [who bor/Peter borrowed a football from]. (23) Da ist der Junge, [der/dem Paul … die Mütze weggenommen hat]. These data suggest that the bulk of the conversion errors did not result from insufficient grammatical knowledge. The children converted object and adverbial relatives to subject relatives although they were able to produce them correctly. But how then do we account for the frequent occurrence of this type of error?

3.3 Discussion Diessel and Tomasello (2005) argue that the conversion errors occurred because subject relatives are more easily activated than other types of relative clauses. One factor determining the ease of activation is frequency of occurrence: Frequently used construction are deeply entrenched in memory and therefore easily activated in language use (cf. Bybee 2006; Bybee & Hopper 2001). However, although subject relatives are among the earliest relative clauses children produce (cf. Diessel 2004: chap 6; Brandt et al. 2008), in spoken discourse they do not seem to be more frequent than direct object relatives (cf. Fox & Thompson 1990; Roland et al. 2007). In fact, Diessel (2004: 146) reports that 4-to-5 year olds produce object relatives more frequently than subject relatives. Moreover, he notes that object relatives are dominant in the ambient language: 57.9 percent of the mothers’ relative clauses in his data are direct object relatives, 34.3 percent are

2. There were also a few subject relatives that were converted to object relatives, but this type of mistake was rare.

Role of frequency and similarity 

subject relatives, and 7.9 percent are adverbial relatives; indirect object relatives and genitive relatives did not appear. Similar frequency distributions have been observed in the ambient language of a German-speaking boy (Brandt et al. 2008) suggesting that input frequency alone does not explain why subject relatives are so easily activated. But what then accounts for the ease of activation? Diessel and Tomasello suggest that children activate subject relatives more easily than other types of relative clauses because subject relatives have the same word order as simple sentences. As can be seen in (24) to (26), in subject relatives the subject precedes all other semantic roles, whereas in object and adverbial relatives it follows the patient or some other semantic role. (24) The man (agent) who opened the door (patient).

subj-relative

(25) The cat (patient) the dog (agent) chased around the garden.

obj-relative

(26) The doctor (goal) the patient (agent) went to last night.

adv-relative

This does not explain why intransitive subject relatives caused fewer errors than transitive subject relatives, but it does provide an explanation for children’s good performance on subject relatives, which has also been observed in many other studies (Smith 1974; Tavakolian 1977; de Villiers et al. 1979; Hamburger & Crain 1982; Kidd & Bavin 2002). That similarity with simple sentences plays an important role in the acquisition of relative clauses has also been proposed by Bever (1970; see also de Villiers et al. 1979), but in contrast to this work, Diessel and Tomasello argue that similarity is not only important to account for children’s ease with subject relatives, it also plays an important role in the acquisition of other structural types of relative clauses. Specifically, they argue that relative clauses constitute a family of constructions that children acquire in an incremental fashion such that new relative clauses are learned based on structures the child already knows. To begin, Diessel and Tomasello argue that the English-speaking children basically produced the same amount of errors in response to direct object relatives, indirect object relatives, and adverbial relatives because these three types of relative clauses involve the same word order. As can be seen in (27), direct object relatives, indirect object relatives, and adverbial relatives include the same sequence of constituents (i.e., … NP NP V …), which contrasts with the constituent order in subject and genitive relatives. This explains why the three types of relative clauses basically caused the same amount of errors although direct object relatives are much more frequent than adverbial relatives and indirect object relatives. In fact, the latter are so rare in the ambient language that children have very little experience with this type of relative clause. However, since indirect object relatives are structurally similar to direct object relatives they did not cause significantly more problems.

 Holger Diessel

(27)

the N [who saw NP] the N [who NP saw] the N [who NP gave NP to] the N [who NP played with] the N [[whose N] chased NP]

subject direct object indirect object adverbial genitive

Note that in German object and adverbial relatives do not form a natural class. Each relative clause is marked by a different case form of the relative pronoun; that is, there are no structural similarities between object and adverbial relatives that could have affected the children’s responses. However, since direct object relatives are more frequent than indirect object relatives and adverbial relatives, the German-speaking children had significantly fewer problems with direct object relatives than with the two other types of relative clauses. Note, however, that although adverbial relatives are more frequent in the ambient language than indirect object relatives (cf. Brandt et al. 2008), they caused significantly more errors in the German study. Diessel and Tomasello suggest that the German-speaking children had particular difficulties with adverbial relatives because these relative clauses are structurally very different from all other types of relative clauses in German. As can be seen in (28), they include a preposition before the relative pronoun whereas all other relative clauses, including indirect object relatives, begin with the relative pronoun. (28)

der Mann, der … der Mann, den … der Mann, dem … der Mann, mit/von dem … der Mann, dessen N …

subject direct object indirect object adverbial genitive

Finally, we have to ask why genitive relatives were almost always incorrect. One of the reasons why children had great difficulties with genitive relatives may be that genitive relatives are extremely rare in the ambient language; but input frequency alone cannot account for children’s poor performance on genitive relatives because indirect object relatives caused significantly fewer problems than genitive relatives despite the fact that both types of relative clauses are basically absent from the ambient language. Both genitive and indirect object relatives are extremely rare in the input, but children had fewer difficulties with indirect object relatives than with genitive relatives because genitive relatives are very different from all other types of relative clauses: They establish the link between main and relative clauses through a genitive attribute that is coreferential with the noun modified by the relative clause and semantically associated with the subsequent noun. In sum, similarity played a key role in this study: • Subject relatives caused fewer problems than other types of relative clauses because they have the same the word order as simple sentences.

Role of frequency and similarity 

• Direct object relatives, indirect object relatives, and adverbial relatives caused basi• •

cally the same amount of problems in the English study because they share important word order properties with each other. Indirect object relatives caused fewer problems than genitive relatives despite the fact that both types of relative clauses are basically absent from the ambient language because indirect object relatives are similar to other types of relative clauses. And genitive relatives and German adverbial relatives caused by far the greatest problems because they are very different from all other types of relative clauses.

Why is similarity such an important factor? It is important because relative clauses are grammatical constructions, i.e., form-function pairings, that are related to each other in an associative network like lexical expressions (cf. Goldberg 1995, 2006). Children acquire this network in a piecemeal, bottom-up fashion, relating new relative clauses to constructions they already know. The development starts with subject relatives in copular constructions, which are only little different from simple sentences—they contain a single proposition and involve the same word order as simple main clauses— and it ends with genitive relatives that are most distinct from all other types of relative clauses. Inspired by this research, Fitz and Chang (2008) conducted a connectionist study in which a recurrent localist network (cf. Elman 1990) was trained to learn various types of relative clauses from a training sample of simple and complex sentences. Interestingly, the model performed in the same way as the children in the Diessel and Tomasello study. One of the factors determining the simulation was input frequency; but in addition, the network’s performance was affected by the similarity between constructions. Manipulating the sentences in the training sample, Fitz and Chang observed that the network’s performance on relative clauses varied with the types of simple (and complex) sentences to which the model was exposed, suggesting that the emergence of a particular type of relative clause is determined by its similarity to simple sentences and other types of relative clauses. Specifically, Fitz and Chang argued that it is the frequent occurrence of the fragment THAT VERB as opposed to THAT ARTICLE NOUN that facilitated the emergence of subject relatives.

4. Study 2 The first study has shown that structural similarity is an important determinant for the processing and acquisition of relative clauses. Specifically, we have seen that children have fewer difficulties with subject relatives than with object and adverbial relatives because they have the same word order as simple sentences, whereas object and adverbial

 Holger Diessel

relatives deviate from the SV(O) schema of ordinary (in)transitive clauses. However, a number of recent studies have shown that the processing load of object relatives is crucially affected by semantic and pragmatic aspects that have been ignored in earlier research on the processing and acquisition of relative clauses. Two findings are relevant. First, several studies have demonstrated that the semantic feature of animacy is an important determinant for the comprehension of relative clauses in adult language (cf. Trueswell et al. 1994; Traxler et al. 2002, 2005; Mak et al. 2002, 2006; Gennari & MacDonald 2008). For instance, Mak et al. (2002) conducted a reading time experiment with Dutch-speaking adults in which animacy had a differential effect on the processing of subject and object relatives. Using various combinations of animate and inanimate nouns, they found no significant difference in reading times between subject and object relatives if the subject of the relative clause is animate and the object inanimate (cf. examples 29–30); it was only when both subject and object relatives were animate that object relatives caused longer reading times than subject relatives (cf. examples 31–32) (cf. Trueswell et al. 1994; Traxler et al. 2002; Mak et al. 2006; Gennari & MacDonald 2008). (29) The burglars who stole the computer …

animate-inanimate/subject-rel

(30) The computer that the burglars robbed … animate-inanimate/object-rel (31) The burglars who robbed the occupant… animate-animate/subject-rel (32) The occupant who the burglars robbed … animate-animate/object-rel*** Second, several experimental studies have shown that the processing difficulty of an object relative is affected by the type of subject it includes. For instance, Warren and Gibson (2002) found that object relatives including a first or second person pronoun as subject (cf. example 33) have shorter reading times than object relatives including a proper name (example 34), which in turn cause shorter reading times than object relatives including a lexical subject (example 35), especially when the subject is indefinite (cf. example 36). Warren and Gibson argue that the NP type of the subject influences the processing of object relatives because it correlates with the accessibility of the referent (cf. Ariel 1990; see also Givón 1983). Other things being equal, the higher the subject on the accessibility scale, the lower the processing load of the relative clause (cf. Warren & Gibson 2005; see Gordon et al. 2001, 2004 and Reali & Christiansen 2007 for somewhat different explanations). (33) The man I met.

First person subject

(34) The man Peter met.

Third person subject

(35) The man the woman met. Definite lexical subject (36) The man a woman met.

Indefinite lexical subject

Role of frequency and similarity 

Continuing this line of research, Kidd et al. (2007) conducted a sentence repetition experiment with 4-to-5 year old English- and German-speaking children in which they manipulated the animacy of the head noun and the NP-type of the subject in object relatives. In accordance with the studies in adult psycholinguistics, they found that an inanimate head and a pronominal subject reduce children’s difficulties with object relatives. Like adult speakers, children find object relatives as easy (or difficult) to comprehend as subject relatives if they modify an inanimate noun and include a pronominal subject. In what follows I will take a closer look at the meaning and function subject and non-subject relatives in spontaneous child language. The study shows that subject relatives comprise a variety of constructions including transitive, intransitive, and copular verbs that occur with various constellations of animate and inanimate nouns, whereas non-subject relatives are realized by prototypical transitive clauses. They usually contain a dynamic verb denoting a goal-directed activity and include a first or second person pronoun as subject functioning as agent or experiencer. It is argued that the prototypical meaning of non-subject relatives helps the child to bootstrap into this type of relative clause.

4.1 Data and coding The investigation is based on observational data from two English-speaking children from the CHILDES database (MacWhinney 2000), Adam (Brown 1973) and Abe (Kuczaj 1976). Adam’s data comprise 55 one-hour recordings that occurred at regular intervals of 2 to 3 weeks between the ages of 2;3 and 5;2; overall there are 180 finite relative clauses in Adam’s corpus. Abe’s data consist of 210 files containing transcripts of 30 minutes recordings that were collected every week between the ages of 2;4 and 5;0; overall Abe’s corpus includes 309 finite relative clauses. In Diessel (2004), I also looked at the relative clauses of four other children, Sarah (Brown 1973), Peter (Bloom 1973), Nina (Suppes 1973), and Naomi (Sachs 1983); but since the corpora of these children are too small to investigate the interaction between semantic and syntactic features, they were excluded from the current investigation.

Table 1. Data. Child

Age

rel-clauses

Adam Abe

2;3–5;2 2;4–5;0

180 309

Recordings 55 one-hour recordings 210 half-hour recordings

 Holger Diessel

The data were coded for various syntactic and semantic variables concerning aspects of both the main clause and the relative clause. The variables of the main clause include: 1. 2. 3. 4.

the meaning of the verb the syntactic function of the head noun the thematic role of the head noun the animacy of the head noun

The variables of the relative clause include: 1. 2. 3. 4. 5.

the meaning of the verb the syntactic function of the relativized role the thematic role of the relativized role the animacy of subject and object the NP type of subject and object

The levels of the variables will be defined in course of the analysis. In order to ensure coding reliability, 100 sentences were double coded by the author and a student assistant. For the syntactic variables, inter-coder reliability was almost 100 percent; for the semantic variables it was between 94.0 percent (thematic role) and 98.0 percent (animacy).

4.2 The animacy of the head and the relativized role In a first step I examined the relationship between the animacy features of the head and the relativized syntactic role. The latter were divided into two basic types, subject relatives, i.e., relative clauses in which the subject is relativized, and non-subject relatives, i.e., relative clauses in which the direct object or an adverbial is relativized; indirect object relatives and genitive relatives did not occur in the data. Interestingly, although children have often difficulties with non-subject relatives in comprehension experiments, both children began to use non-subject relatives at around the same time as subject relatives and eventually produced more non-subject relatives than subject relatives: 64.5 percent (n = 116) of Adam’s relative clauses and 57.3 percent (n = 177) of Abe’s relative clauses are non-subject relatives. The head of the relative clause was also divided into two basic types: animate head nouns denoting humans and animals, including toy animals, and inanimate head nouns denoting concrete objects and abstract entities. The vast majority of the children’s relative clauses elaborate an inanimate noun: 78.3 percent (n = 141) of Adam’s relatives and 79.0 percent (n = 244) of Abe’s relatives are attached to an inanimate noun. Figure 2 shows that the animacy of the head correlates with the relativized syntactic role. Subject relatives are common with both animate and inanimate nouns—there is only a small difference between them. On average, 42.4 percent of the children’s subject

Role of frequency and similarity  subject relative

non-subject relative 8,30%

42,40% 57,60% 91,70%

animate head inanimate head Figure 2. Animate and inanimate heads of subject and non-subject relatives.

relatives occur with animate nouns and 57.6 percent occur with inanimate nouns. However, non-subject relatives occur predominantly with inanimate nouns. As can be seen in Figure 2, on average 91.7 percent of Adam’s and Abe’s non-subject relatives are attached to an inanimate noun and only 8.3 percent occur with an animate noun. A 2 x 2 χ2-analysis revealed a significant association between the animacy of the head and the syntactic function of the relativized role in the data of both children (Abe: χ2 = 52.67, df = 1, p < 0.001, φ = 0.418; Adam: χ2 = 27.09, df = 1, p < 0.001, φ = 0.393) suggesting that the semantic feature of animacy is an important aspect of children’s subject and non-subject relatives. But why does the animacy of the head correlate with the relativized role? The answer appears to be straightforward: The correlation reflects the prototypical links between syntactic and semantic roles. Since animate referents are likely to function as agents, they are commonly modified by subject relatives, i.e., relative clauses in which the animate head functions as an agent or experiencer, and since inanimate referents are likely to function as patients, they are commonly modified by non-subject relatives, i.e., relative clauses in which the inanimate head functions as patient, theme, or location. In other words, I suggest that the semantic biases in children’s subject and non-subject relatives arise from entrenched relationships between semantic and syntactic roles. One can think of these relationships as an associative network that emerges from children’s experience with verb-argument constructions: The more often a semantic role is expressed by a particular syntactic category, the stronger the associative link between form and meaning. This explains why object relatives cause fewer difficulties in comprehension studies if they modify an inanimate noun rather than an animate noun. However, it

 Holger Diessel

does not fully explain the animacy features of subject relatives: While subject relatives are more frequent with animate heads than non-subject relatives, most of the children’s subject relatives modify inanimate nouns (Figure 2), suggesting that subject relatives do not instantiate the prototypical links between syntactic and semantic roles. In what follows I show that subject relatives comprise a variety of constructions that vary in terms of their semantic and pragmatic features, whereas non-subject relatives are commonly expressed by prototypical transitive clauses involving an animate agentive subject, a dynamic verb, and an inanimate object. Based on these data, I argue that the prototypical meaning of non-subject relatives helps the child to process (and learn) the particular structural properties of non-subject relatives.

4.3 The NP-type of the subject The second aspect of non-subject relatives that has been studied intensively in the recent psycholinguistic literature on relative clauses is the NP-type of the subject (see above). In accordance with the Kidd et al. study, I found that non-subject relatives typically include a pronominal subject, notably a first and second person pronoun is very common. As can be seen in Figure 3, an average of 80.4 percent of Adam’s and Abe’s non-subject relatives include I, you, or we as subject (cf. Fox & Thompson 1990, 2007). The rest occur with third person pronouns (mean 7.7 percent), definite NPs (mean 10.0 percent), and a few indefinite nouns (mean 1.9 percent). While the occurrence of the various NP-types is strongly skewed in favour of pronominal subjects, it must be emphasized that the frequency distribution of the 80,0% 70,0% 60,0% 50,0% 40,0%

80,4%

30,0% 20,0% 10,0% I/you/we

7,7%

10,0%

1,9%

he/she/it/they

definite NP

Indefinite NP

Figure 3. Subjects of non-subject relatives.

Role of frequency and similarity 

various NP-types in non-subject relatives corresponds very closely with the distribution of the various NP-types in simple transitive clauses. Examining a random sample of 350 transitive (main) clauses from Adam and Abe’s corpus I found that an average 86.9 percent of all subjects are first or second person pronouns, 7.5 percent are third person pronouns, 3.7 percent are definite NPs, and only 1.8 percent are indefinite NPs. In other words, the predominant use of pronominal subjects is not a particular trait of non-subject relatives but is characteristic of ordinary (i.e., prototypical) transitive clauses. By contrast, the subjects of subject relatives are radically different from the subjects of simple (in)transitive clauses. Since relative clauses do not modify first or second pronouns, they are exclusively used with third person subjects. Moreover, the subject of a subject relative is often a new discourse referent. As can be seen in Figure 4, the vast majority of Adam and Abe’s subject relatives are attached to a lexical noun functioning as subject inside of the relative clause. Note that more than half of the head nouns are indefinite, expressed by an indefinite lexical NP or an indefinite pronoun, introducing a new referent that functions as the subject of the relative clause. Thus, the two types of relative clauses occur with very different types of subjects: non-subject relatives occur with highly accessible subjects that are characteristic of prototypical transitive clauses, whereas the subjects of subject relatives are third person referents that are often newly introduced in the preceding main clause.

40,00%

30,00%

43,30%

20,00%

31,70% 19,70%

10,00% 5,40% this/that

definite NP

Figure 4. Subjects of subject relatives.

indefinite NP

indefinite PRO

 Holger Diessel

4.4 The verb of the relative clause Subject and non-subject relatives do not only occur with different subjects, they also occur with different types of verbs. The vast majority of the children’s non-subject relatives include a transitive verb denoting a physical or cognitive activity (cf. examples 37–39); there are only a few intransitive non-subject relatives in the data in which an adverbial is relativized (cf. example 40). (37) Where’s the balloon [I made]?

Abe 3;1

(38) Dose are bugs [that I throw].

Adam 3;7

(39) No the one [you found …].

Abe 3;9

(40) The inside of it [where we sit].

Abe 3;6

In contrast, subject relatives occur with both transitive and intransitive verbs. In fact, the majority of the children’s subject relatives contain an intransitive verb or a copula, notably the copula be. As can be seen in Figure 5, an average of only 37.6 percent of the children’s subject relatives are transitive, the rest include either an intransitive verb (43.8 percent) or a copula (18.6 percent), whereas non-subject relatives are mostly transitive: 95.1 percent of the children’s non-subject relatives are transitive and only 4.9 percent occur with intransitive and copular verbs. Interestingly, transitive and intransitive subject relatives are headed by different semantic types of nouns. Most of the copular and intransitive subject relatives are subject relative

non-subject relative 3,50%

18,60%

1,40%

37,60%

95,10%

43,80%

transitive intransitive copula Figure 5. Verbs of subject and non-subject relatives.

Role of frequency and similarity 

headed by inanimate nouns: On average 71.7 percent of the children’s intransitive subject relatives and 52.9 percent of their copular subject relatives modify an inanimate noun (cf. Figure 6). Note that the intransitive subject relatives comprise unergative and unaccusative verbs (including transitive verbs in passive voice), which tend to occur with different semantic types of nouns: Unergative verbs occur with both animate and inanimate nouns (cf. examples 41 and 42), whereas unaccusative verbs (and passivized transitive verbs) are almost exclusively used with inanimate nouns (cf. examples 43). (41) The doggie [that runs away].

[Adam 3;8]

(42) Look at that big truck [(dat) going some place].

[Adam 3;0]

(43) The wheel [that’s broken].

[Abe 3;10]

Transitive subject relatives are more frequent with animate head nouns than intransitive and copular subject relatives. As can be seen in Figure 6, an average of 63.7 percent of the transitive subject relatives are attached to an animate noun and only 36.4 percent modify an inanimate noun. The difference between the three types of subject relatives is highly significant in the data of both children (Abe: χ2 = 14.53, df = 2, p < 0.001, φ = 0.336; Adam: χ2 = 10.49, df = 2, p < 0.001, φ = 0.441). Note, however, that the children’s (transitive) non-subject relatives include a larger proportion of animate subjects than the children’s transitive subject relatives. Averaged across the two children, 93.2 percent of Adam and Abe’s non-subject relatives contain an animate subject, which corresponds quite closely to the proportion of animate subjects in the children’s simple transitive clauses (96.1 percent); whereas only 63.7 percent of the transitive subject relatives occur with an animate subject. The difference between

100,00%

75,00%

52,90%

36,40%

71,70%

inanimate animate

50,00%

25,00%

47,20%

63,70%

28,40% 0,00%

intransitive subject REL

copular subject REL

transitive subject REL

Figure 6. Animate and inanimate heads of transitive, copular, and intransitive subject relatives.

 Holger Diessel

transitive subject and (transitive) non-subject relatives is highly significant (Abe: χ2 = 43.60, df = 1, p < 0.001, φ = 0.435; Adam: χ2 = 26.1, df = 1, p < 0.001, φ = 0.440) suggesting that object and adverbial relatives include more prototypical subjects than subject relatives. In order to test this hypothesis, I analyzed the thematic roles of the subjects in both subject and non-subject relatives, using the following definitions of semantic roles (cf. Huddleston & Pullum 2002): 1. 2. 3. 4. 5. 6. 7. 8.

agent (instigator of a physical activity) patient (animate or inanimate being affected by a physical activity) experiencer (animate being experiencing or perceiving an event) recipient (animate being receiving an entity) beneficary (animate being benefiting from an activity) instrument (inanimate entity used to accomplish a goal) location (place, destination, source) theme (entity that is perceived, known, located, or possessed).

In accordance with the above hypothesis, I found that an average of 88.7 percent of Adam’s and Abe’s non-subject relatives include an agent or experiencer as subject. In contrast, subject relatives include a much smaller proportion of agentive subjects: Only 48.1 percent of the children’s subject relatives include a prototypical subject functioning as agent or experiencer of the activity expressed by the verb; the remaining 51.9 percent function as patient, theme, or location. This is of course partly due to the fact that the children’s subject relatives include many intransitive and copular verbs, but even if we restrict the analysis to transitive verbs an average of only 55.2 percent of Adam and Abe’s subject relatives occur with an agent or experiencer as subject. In particular, experiencer subjects are extremely rare in subject relatives. Overall, there are only 5 subject relatives in the entire database in which the subject of a transitive subject relative clause functions as an experiencer, i.e., the mental agent of a cognitive or psychological state or activity. This is reflected in the types of verbs that subject and non-subject relatives include. Concentrating on relative clauses with transitive verbs, I distinguished between three different semantic verb types: 1. Verbs denoting a physical activity (e.g., put, make) 2. Verbs denoting cognition, perception, or communication (e.g., know, see, say) 3. Verbs denoting a state or possession (e.g., be, have, belong) While this classification surely has its problems, Figure 7 shows that there are some striking differences between the two types of relative clauses: Both subject and nonsubject relatives are commonly used with (transitive) activity verbs, but while nonsubject relatives are also commonly used with verbs of cognition, perception, or

Role of frequency and similarity  subject relative

non-subject relative 11,10%

45,00%

51,50%

29,80%

59,10%

3,50% activity cognition/perception/communication state/possession Figure 7. The meaning of transitive verbs in subject and non-subject relative..

communication, subject relatives are only rarely used with these verbs, including instead many stative verbs and verbs denoting possession; in particular, the verb have is very frequent in transitive subject relatives. The difference between subject and nonsubject relatives is again highly significant (Abe: χ2 = 82.38, df = 2, p < 0.001, φ = 0.519; Adam: χ2 = 27.46, df = 2, p < 0.001, φ = 0.391).

4.5 The nominal referents of the relative clause Finally, I examined the semantic differences between subject, object, and adverbials in the children’s relative clauses with multiple referents. Disregarding intransitive subject relatives with a single referent (i.e., subject relatives including only a subject), I distinguished between four different types of relative clauses: 1. Relative clauses including an animate subject and an inanimate object (or adverbial).3 2. Relative clauses including an animate subject and an animate object (or adverbial).

3. In relatives clauses with more than two referents I concentrated on the core roles, i.e., subject and object, and disregarded adverbials.

 Holger Diessel

3. Relative clauses including an inanimate subject and an inanimate object (or adverbial). 4. Relative clauses including an inanimate subject and an animate object (or adverbial). Figure 8 shows that while subject relatives occur with various combinations of animate and inanimate nouns, non-subject relatives typically include an animate subject and an inanimate object or adverbial. An average of 44.9 percent of the children’s subject relatives include an animate subject and an inanimate object (or adverbial) as in prototypical transitive clauses; but very often the two referents are both animate or both inanimate: 14.6 percent of the children’s subject relatives occur with two animate nouns and 36.8 percent occur with two inanimate nouns; that is, in more than half of the children’s subject relatives, subject and object/adverbial do not differ in terms of their animacy features. By contrast, non-subject relatives are strongly skewed in favour one particular type. As can be seen in Figure 8, 88.6 percent of the children’s non-subject relatives contain an animate subject and an inanimate object (or adverbial), all other types are infrequent, supporting the hypothesis that non-subject relatives are prototypical transitive clauses, whereas subject relatives are distributed across various types of constructions that are low on the transitivity scale. subject relative 3,70%

36,80%

non-subject relative 3,10%

8,30%

44,90% 88,60% 14,60%

animate SUBJ - inanimate OBJ/ADV animate SUBJ - animate OBJ/ADV inanimate SUBJ - inanimate OBJ/ADV inanimate SUBJ - animate OBJ/ADV Figure 8. Animate and inanimate NPs in subject and non-subject relatives.

Role of frequency and similarity 

5. Conclusion In the usage-based approach to language acquisition, grammatical development reflects the child’s experience with language. But what exactly constitutes the child’s linguistic experience? Two aspects have been emphasized in recent studies on the acquisition of morphology and the emergence of item-based constructions: Frequency of occurrence, which determines the level of entrenchment, and structural similarity, which underlies the emergence of general grammatical patterns (or schemas). This study has shown that frequency and similarity are not only crucial for the early stages of grammatical development but are equally important for the acquisition of more complex grammatical patterns such as relative clauses. Specifically, the paper has argued that the acquisition of relative clauses is influenced by children’s prior knowledge of simple sentences. The first study has shown that in comprehension, children have often fewer difficulties with subject relatives than with object, adverbial, and genitive relatives because subject relatives involve the same sequence of subject, verb, and object as simple sentences whereas non-subject relatives deviate from the familiar SV(O) pattern. Moreover, the study suggests that structural similarities between object, adverbial, and genitive relatives affect the processing and development of these constructions in the early stages of grammatical development. The second study has shown that although non-subject relatives are structurally distinct from ordinary SV(O) clauses, semantically they are expressed by prototypical transitive clauses, whereas subject relatives comprise a variety of constructions including transitive, intransitive, and copular verbs. Interestingly, although subject relatives have the same word order as ordinary SV(O) clauses, semantically they often deviate from ordinary (in)transitive clauses. For instance, while the vast majority of the children’s simple SV(O) clauses include a pronominal subject, notably a first or second person pronoun, subject relatives are exclusively used with third person subjects that are often newly introduced into the discourse. In general, subject relatives comprise a diverse group of constructions with rather unusual semantic and pragmatic properties, whereas non-subject relatives are commonly expressed by prototypical transitive constructions including a pronominal animate subject functioning as agent or experiencer of a dynamic verb with an inanimate object functioning as patient or theme. The semantic and pragmatic properties of non-subject relatives are thus predictable from the properties of prototypical transitive clauses, which arguably may help the child to bootstrap into this type of relative clause. In other words, the difficulties that arise from the unusual structure of non-subject relatives are mitigated by the fact that non-subject relatives instantiate the semantic and pragmatic properties of prototypical transitive clauses.

 Holger Diessel

References Andrews, A.D. 2007. Relative clauses. In Language Typology and Syntactic Description, 2nd Edn. T. Shopen (Ed.), 206–236. Cambridge: CUP. Ariel, M. 1990. Accessing Noun-Phrase Antecedents. London: Routledge. Bever, T.G. 1970. The cognitive basis for linguistic structures. In Cognition and Development of Language, J.R. Hayes (Ed.), 279–352. New York NY: Wiley. Bloom, L.P. 1973. One Word at a Time. The Use of Single Word Utterances. The Hague: Mouton. Brandt, S., Diessel, H. & Tomasello, M. 2008. The acquisition of German relative clauses. A case study. Journal of Child Language 34: 1–24. Brown, R. 1973. A First Language, Cambridge MA: Harvard University Press. Bybee, J. 2006. From usage to grammar: The mind’s response to repetition. Language 82: 711–733. Bybee, J. & Slobin, D. 1982. Rules and schemas in the development and use of the English past tense. Language 58: 265–289. Bybee, J. & Hopper, P. 2001. Frequency and the Emergence of Linguistic Structure. Amsterdam: John Benjamins Clancy, P., Lee, H. & Zoh, M.-H. 1986. Processing strategies in the acquisition of relative clauses: Universal principles and language-specific realizations. Cognition 24: 225–262. Corrêa, L.M.S. 1995. An alternative assessment of children’s comprehension of relative clauses. Journal of Psycholinguistic Research 24: 183–203. Dasinger, L. & Toupin, C. 1994. The development of relative clause functions in narratives. In Relating Events in Narrative: A Cross-linguistic Developmental Study, R.A. Berman & D.I. Slobin (Eds), 457–514. Hillsdale NY: Lawrence Erlbaum. de Villiers, J.G., Tager Flusberg, H.B., Hakuta, K. & Cohen, M. 1979. Children’s comprehension of relative clauses. Journal of Psycholinguistic Research 8: 499–518. Diessel, H. 2004. The Acquisition of Complex Sentences. Cambridge: CUP. Diessel, H. 2007a. Frequency effects in language acquisition, language use, and diachronic change. New Ideas in Psychology 25: 108–127. Diessel, H. 2007b. A construction-based analysis of the acquisition of East Asian relative clauses. Studies in Second Language Acquisition 29: 311–320. Diessel, H. & Tomasello, M. 2000. The development of relative clauses in spontaneous child speech. Cognitive Linguistics 11: 131–151. Diessel, H. & Tomasello, M. 2005. A new look at the acquisition of relative clauses. Language 81: 1–25. Elman, J.L. 1990. Finding structure in time. Cognitive Science 14: 179–211. Fitz, H. & Chang, F. 2008. The role of the input in a connectionist model of the accessibility hierarchy in development. In Proceedings of the 32nd Boston University Conference on Language Development, 120–131. Somerville MA: Cascadilla Press. Fox, B.A. & Thompson, S.A. 1990. A discourse explanation of the grammar of relative clauses in English conversation. Language 66: 297–316. Fox, B.A. & Thompson, S.A. 2007. Relative clauses in English conversations. Studies in Language 31: 293–326. Gennari, S.P. & MacDonald, M.C. 2008. Semantic indeterminacy in object relative clauses. Journal of Memory and Language 58: 161–187. Gentner, D. 1983. Structure-mapping: A theoretical framework for analogy. Cognitive Science 7: 155–170. Gentner, D. & Median, J. 1998. Similarity and the development of rules. Cognition 65: 263–262.

Role of frequency and similarity 

Gibson, E. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition 68: 1–76. Givón, T. (Ed.). 1983. Topic Continuity in Discourse: Quantified Cross-Language Studies. Amsterdam: John Benjamins. Givón, T. 1990. Syntax: A Functional-typological Introduction, Vol. II. Amsterdam: John Benjamins. Givón, T. 2008. The ontogeny of relative clauses: How children learn to negotiate complex reference. In The Genesis of Syntactic Complexity, T. Givón, Ch. 8. Amsterdam: John Benjamins. Goodluck, H. & Tavakolian, S. 1982. Competence and processing in children’s grammar of relative clauses. Cognition 11: 1–27. Goldberg, A.E. 1995. A Construction Grammar Approach to Argument Structure. Chicago IL: University of Chicago Press. Goldberg, A.E. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: OUP. Gorden, P.C., Hendrick, R. & Johnson, M. 2001. Memory interference during language processing. Journal of Experimental Psychology: Learning, Memory and Cognition 27: 1411–1423. Gorden, P.C., Hendrick, R. & Johnson, M. 2004. Effects of noun phrase type on sentence complexity. Journal of Memory and Language 51: 97–114. Hamburger, H. & Crain, S. 1982. Relative Acquisition. In Language Development, Vol. I, Syntax and Semantics, S. Kuczaj (Ed.), 245–274. Hillsdale NY: Lawrence Erlbaum. Hawkins, J.A. 1987. Implicational universals as predictors of language acquisition. Linguistics 25: 453–473. Hawkins, J.A. 1994. A Performance Theory of Order and Constituency. Cambridge: CUP. Hermon, G. 2004. The acquisition of Indonesian relative clauses. Paper presented at the Workshop on Typology, Acquisition and Processing of Relative Clauses, Max Plank Institute for Evolutionary Anthropology, Leipzig. Hofstadter, D. 2001. Analogy as the core of cognition. In The Analogical Mind: Perspectives from Cognitive Science, D. Gentner, K.J. Holyoak, & B.N. Kokinov (Eds), 499–538. Cambridge MA: The MIT Press/Bradford Book. Huddleston, R. & Pullum, G.K. 2002. The Cambridge Grammar of the English Language. Cambridge: CUP. Hudelot, C. 1980. Qu’est-ce que la complexité syntaxique? L’ exemple de la relative. La Linguistique 16: 5–41. Keenan, E. & Hawkins, S. 1987. The psychological validity of the accessibility hierarchy. In

Universal Grammar: 15 Essays, E. Keenan (Ed.), 60–85. London: Croom Helm.

Kidd, E. & Bavin, E.L. 2002. English-speaking children’s comprehension of relative clauses: evidence for general-cognitive and language-specific constraints on development. Journal of Psycholinguistic Research 31: 599–617. Kidd, E., Brandt, S., Lieven, E. & Tomasello, M. 2007. Object relatives made easy: A crosslinguistic comparison of the constraints influencing young children’s processing of relative clauses. Language and Cognitive Processes 22: 860–897. Kim, Y.-J. 1989. Theoretical implications of complement structure acquisition in Korean. Journal of Child Language 16: 573–598. Kuczaj, S. 1976. -ing, -s and -ed: A study of the acquisition of certain verb inflections. Ph.D. dissertation, University of Minnesota. Lambrecht, K. 1988. There was a farmer had a dog: Syntactic amalgams revisited. Proceedings of the Annual Meeting of the Berkeley Linguistics Society 14: 319–339.

 Holger Diessel Lieven, E., Behrens, H., Speares, J. & Tomasello, M. 2003. Early syntactic creativity: A usagebased approach. Journal of Child Language 30: 333–370. MacWhinney. B. 2000. The CHILDES Project: Tools for Analyzing Talk, Vol. 2, The Database. Hillsdale NJ: Lawrence Erlbaum. Mak, W.M., Vonk, W. & Schriefers, H. 2002. The influence of animacy on relative clause processing. Journal of Memory and Language 74: 50–68. Mak, W.M., Vonk, W. & Schriefers, H. 2006. Animacy in processing relative clauses. Journal of Memory and Language 54: 466–490. O’Grady, W. 1997. Syntactic Development. Chicago IL: The University of Chicago Press. Ozeki, H. & Shirai, Y. 2005. Semantic bias in the acquisition of relative clauses in Japanese. In Proceedings of the 29nd Annual Boston University Conference on Language Development, Vol. II, 459–470. Somerville MA: Cascadilla Press. Reali, F. & Christiansen, M.H. 2007. Processing of relative clauses is made easier by frequency of occurrence. Journal of Memory and Language 57: 1–23. Roland, D., Dick, F. & Elman, J.L. 2007. Frequency of basic English grammatical structures: A corpus analysis. Journal of Memory and Language 57: 348–379. Sachs, J. 1983. Talking about there and then: The emergence of displaced reference in parentchild discourse. In Children’s Language, Vol. IV, K.E. Nelson (Ed.), 359–438, Hillsdale NJ: Lawrence Erlbaum. Sheldon, A. 1974. The role of parallel function in the acquisition of relative clauses in English. Journal of Verbal Learning and Verbal Behaviour 13: 272–281. Smith, M.D. 1974. Relative clause formation between 29–36 months: A preliminary report. Stanford Papers and Reports on Child Language Development 8: 104–110. Suppes, P. 1973. The semantics of children’s language. American Psychologist 88: 103–114. Tavakolian, S.L. 1977. Structural Principles in the Acquisition of Complex Sentences. Ph.D. dissertation, University of Massachusetts. Tomasello, M. 2000. Do young children have adult syntactic competence? Cognition 74: 209–253. Tomasello, M. 2003. Constructing a Language. A Usage-based Theory of Language Acquisition. Cambridge MA: Harvard University Press. Traxler, M.J., Morris, R.K. & Seely, R.S. 2002. Processing subject and object relative clauses: Evidence from eye movement. Journal of Memory and Language 74: 69–90. Traxler, M.J., Williams, R.S, Blozis, S.A. & Morris, R.K. 2005. Working Memory, animacy, and verb class in the processing of relative clauses. Journal of Memory and Language 53: 204–224. Trueswell, J.C., Tanenhaus, M.K. & Garnsey, S.M. 1994. Semantic influences on parsing: Use of thematic role information in syntactic ambiguity resolution. Journal of Memory and Language 33: 285–318. Wanner, E. & Maratsos, M. 1978. An ATN approach to comprehension. In Linguistic Theory and Psychological Reality, M. Halle, J. Bresnan & G. Miller (Eds), 119–161. Cambridge: CUP. Warren, T. & Gibson, E. 2002. The influence of referential processing on sentence complexity. Cognition 85: 79–112. Warren, T. & Gibson, E. 2005. Effects of NP type in reading cleft sentences in English. Language and Cognitive Processes 20: 751–767.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish Cecilia Rojas-Nieto

Universidad Nacional Autónoma de México

This paper builds on previous Usage-based accounts of developing sentence complexity (Diessel 2004; Diessel & Tomasello 2000, 2001), considering early relative constructions (RC) in Spanish. RCs development shows various “starting small” processes (Elman 1993): Most RCs show no embedding; they are dialogical co-constructional results or take an absolute position and no intonation integration with the main clause. When embedded, constructional frames are lexically biased with an open slot for Head RC insertion. RCs internal structure is mostly similar to independent clause type, with no gap nor genuine ‘relative’ function for the relative pronoun. In sum, RCs show an exemplar based acquisition and individually preferred constructional frames. All these phenomena point towards a non linear, experience based learning, affected by frequency and function.

1. Presentation Spanish relative constructions (henceforth RCs) are produced with, primarily, a relative pronoun strategy, and are commonly recognized as complex structures.1 They tend to display a set of syntactic features (‘operations’): embedding; head dependence; possibly head anaphoric marking; focusing of an internal constituent; coreference calibration of the focused constituent; coreference marking — by means of a relative pronoun (rel) in an initial position; an internal “missing constituent” gap (in S and O relatives) or a resumptive pronoun (res) (in IO relatives) (Brucart 1999). Given this set of properties, noted across languages, relative constructions have been considered to be a late developmental achievement in child language acquisition (Echeverría 1978; Hurtado 1984). Descriptions of the acquisition of RCs by children tend to emphasize some default strategy, selected among possible options in terms of simplicity, markedness or lower

. Gerundive constructions are heavily restricted and not preferred.

 Cecilia Rojas-Nieto

processing cost, together with a linear developmental path going from simpler to complex, unmarked to marked, or accessible to inaccessible. Thus, some studies of RC acquition have based their developmental hypotheses on the relativization accessibility hierarchy: subject > object > oblique > possessive (Keenan & Comrie 1977; Barriga 2002); a processing motivated preference (cf. Prideaux & Baker 1986) to have the same function in the antecedent NP and rel: S[S] and O[O], rather than S[O], or O[S] (Echeverría 1978; Hurtado 1984; Sicuro-Corrêa 1995). Simpler syntactic structure, whether adjoined (Hale 1976) or conjoined (Givón 2008), may be preferred over an embedded one (Tavakolian 1981). Lexically-headed rather than determiner-headed RCs are considered as basic, since various models would take determiner headed RCs to be contextually dependent and/or anaphoric reductions of lexically headed constructions (Bello 1847/1988: §§323–325; Brucart 1999). Despite the appealing and elegant predictions that might point towards a converging default as a starting point for children to develop RCs, it is well known that children’s data on various constructions and categories development have proven resistant to initial defaults (Dabrowska 2001; Gathercole, Sebastián & Soto 1999; Rojas 2003), as well as to abstract syntactic principles (Dabrowska & Lieven 2005; McClure, Pine & Liven 2006; Lieven, Pine & Baldwin 1997; Tomasello 2000; 2003; etc.). It is thus common in recent studies on RC development inspired by Usage Based theories of language acquisition that no default strategy is even mentioned (Diessel 2004).2 According to this view, children start to produce early RCs not necessarily following the abstract syntactic predictions, but rather revealing the effects of experience and familiar use. From this children adopt selective and lexically specific construction frames — i.e., form-function pairs — which despite their apparent complexity are in fact monopropositional in nature and reveal a unitary communicative intent. In his foundational study, Diessel (2004) elaborates extensively on this point. Early RCs are mainly presentative constructions (Here is a rabbit that I’m patting: Diessel 2004: 3), which Diessel considers to be syntactically simple despite their complex appearance, since they correspond to a single assertion and are dedicated to introducing new referents in discourse (Lambrecht 1988; Moreno Cabrera 1999; see also Givón 2008 for a different perspective). The evidence presented there points towards one of the main proposals in Usage Based research: that early RCs are based on concrete exemplars experienced by a child. This usage is, by definition situated, particular and individual, and is the raw material from which the child extracts chunks and pieces — not necessarily

2. Cf. also, Diessel & Tomasello 2001; Diessel & Tomasello 2005; Kidd, Brandt, Lieven & Tomasello 2007; Tomasello 2003.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

atomic — whose possible formants are later analysed and gradually organized. Every child thus finds internal patterns and regularities, builds analogies and relations, and organizes a network among them (Tomasello 2003). We must be aware that these early selected exemplars neither reproduce nor obliterate by definition the abstract regularities supposed to define linguistic facts; but they perform (or expose) those regularities in a probabilistic way, with all haphazards and vagaries that real situated, concrete language displays in particular ecological niches (Givón 2008; in this volume). Indeed, as Givón (2008) emphasizes, early relative constructions involve from the start various pragmatic functions related to referential coherence: reference introduction, reference tracking, and reference negotiacion. RCs thus present various structural means for performing these functions, not necesarilly in embedded frames, nor only in existential-presentatives, but in many functionally-motivated paratactic structures whose antecedents may also be present in parental use. From another perspective, Diessel’s proposal that supposedly complex constructions like the presentative and others, are in fact simple and monopropositional, may be considered as an instance of ‘starting small’. Indeed, evidence has been obtained in other developmental and problem-solving domains that ‘starting small’ may be a way to enter complex systems (Elman 1993; Newport 1990; Seidenberg & MacDonald 1999). In this view, complex tasks may not be detected as complex but reanalyzed as simpler. Once so reduced, they may be solved by undemanding means (Newport 1990; Rojas submitted; Seidenberg & MacDonald 1999). There is thus a plausible case for the early acquisition of RCs and other complex syntactic structures as simple ones. Building upon some aspects of these proposals — Usage-Based, functional motivation, and ‘starting small’ —this study will undertake to analyse the development of RCs in Spanish. Following classical (Bowerman 1979; Braine 1976; Limber 1973), and recent child language studies on clause combining (Diessel 2004; Diessel & Tomasello 2000 and 2001; Givón 2008; in this volume; Rojas submitted), we can argue that children have their own concrete and simple way of developing complex constructions, based on their individual experience. This research thus assumes that, given functional motivation in the domain of reference tracking, (i) children will adopt particular lexical-specific structural frames not for their markedness or complexity properties, but rather in line with experienced use; (ii) with the effect of a percolating complexity through their own processing resources. The focus of this work is the emergence of structure in children’s situated interaction, and the way children solve complex tasks without recourse to abstract exotic strategies. The following research questions will guide our analysis:

• How complex are relative constructions in early child language? • Is there evidence for the initial selection of less complex structures as a default? • Can we trace back ‘starting small’ effects in children’s data?

 Cecilia Rojas-Nieto

2. The data The data to be considered here come from the corpus ETAL: Etapas tempranas en la adquisición del lenguaje (Early stages in language acquisition), from the Instituto de Investigaciones Filológicas, at the Universidad Nacional Autónoma de México (Rojas 2007). Three young subjects are used: two girls and one boy. They are all Spanish monolingual children of urban educated families. The age range is from 4;02 ~ 4;00 (last observation) to 2;3–3;04 (no observed RCs). The usual criteria for data selection have been followed: only spontaneous child-produced constructions have been considered, and no successive reiterations have been counted. The analysis will consider only RCs marked by the relative pronoun que (1i–1v), involving a total of 312 tokens. Other RCs marked by quien, ‘who’, cuyo ‘whose’, are absent from the children’s data. The infrequent cases with other relative markers, like donde ‘where’ (2a) and cuando ‘when’ (2b) or cual ‘which’, with or without a lexical head, have not been considered, and were not counted in data presentation in table 1. Table 1. Data Base. Observations

Child Age range conversational (in months) turns

Time

RCs

Flor (Fem)

31–67

24,318

27 m – 48 m

74 hrs.

188

Elia (Fem)

20–29

7,244

40 m – 50 m

20 hrs.

80

Julio (Male) Total

20–30 57

8,835 40,397

37 m – 48 m 27 m – 50 m

22 hrs. 116 hrs.

44 312

Que-marked RCs may be lexically headed by a noun phrase or a bare noun (1i); they may also be headed by a pronoun (1ii–1iii) or a determiner (1iv–1v). These last –determiner-headed relatives (det rel)– involve a definite determiner marked for gender and number (el, la, lo ‘the m/f/n singular’; los, las ‘the m/f plural’) plus a rel (1iv–1v). Various arguments have been made that det is a derived head or anaphoric trace of the missing lexical head (Brucart 1999). Since this det was historically a demonstrative, it is possible that it will keep some indexical force, so that it need not be anaphoric, but is still deictic. In our analysis and data presentation, we will take det to be an anaphoric/indexical head. But its status in children’s grammar will remain an open question pending further research. (1) i.

una casita que tiene mucho espacio one house-dim rel have-pr.3s much room ‘a little house that has lots of space’.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

ii. ese que tienes that rel have-pr.2s ‘that one that you have’. iii. vi una, una que tenía espuma see-pst.1s one one rel have-impf.3s foam ‘I saw one, one that had foam’. iv. estamos viendo lo que salpica be-pr.1p see-ger det rel splash-pr.3s ‘we are looking at what (=the that) splashes’. v. dame la que tiene puntitos give-imp=io.1s det rel have-pr.3s spot-dim-m-p ‘give me the one that (lit. = the that) has little spots’. (2) i.

Al cuarto a donde se fue to=the room to where rfl.3s=go.pst.3s ‘to the room where Kiso (dog’s name) went’.

Kiso Kiso

ii. Un día cuando estaba pequeña me ponía One day when be-impf.1s small o1s=put-impf.1s ‘One day when I was small I used to put on me (cream)’.

3. Analysis 3.1 Starting points: Default or CRs diversity? Our RC data in (1) and (2) make it clear that individual children show a variety of RC types. What is more, each child prefers different constructional types for her earliest RCs. Just taking into consideration the first five RCs produced by each child, we get a plethora of different constructions. (a) Both: NP-headed and det-headed RCs (3i–3iv vs. 3v). (b) RCs with no internal predication (3i); RCs with an overt internal predication (3ii–3v), which may be ritualized or formulaic (vgr. encontrar ‘find’, for Julio: 4i–4iv–). (c) Relative pronoun (rel) with no internal function (3i), or a function lexically determined by the preferred/ritualized predicate (f.i. encontrar ‘find’ determines an Object rel in 4i–4iv). (d) Embedded RCs may be in a construction with a free head, or inserted in a preferred/ritualized verb frame: like mira ‘look (imperative)’ for Julio (4i–4iv), and éste es ‘this/it is’ for Elia (5i–5iii). In contrast, in Flor’s data, RCs mainly adhere to syntactically free NPs (3i–3iii). Consider the following first five exemplars from every child, which show all these syntactically variegated constructional exemplars.

 Cecilia Rojas-Nieto

(3) flor (2;03–2;04) (M = mother) i. M: ¿qué es eso? what’s that? F: la cama que agua, que…aquí the bed rel water, rel…aquí ‘the bed that water, that … here’. ii. M: F:

qué estas viendo, Flor? ‘what are you looking at, Flor?’ el coche que maneja the car rel drive-pr.3s (=moves) ‘the car that moves’.

iii. F:

ese señor que sí tiene huevo that man rel aff have-pr.3s egg ‘that man that really has (an) egg’.

iv. F: M: F:

esa es la ropa this be.pr.3s det clothes ‘this is the clothes’ ¿cuál? ‘which one?’ que… de ese nene es rel of that baby be.pr.3s ‘which, belongs to that baby’.

v. F:

mía e que se cayó look-imp det rel rfl.3s=fall.down-pst.3s ‘look the one which (lit. the that) fell down’.

(4) Julio (3;01–3;02) i. J: mí(r)a que’ cont(r)ó look-imp rel find-pst.3s ‘look that (=what) he found’. ii. J:

una, mí(r)a, que’ cont(r)é! one, look-imp rel find-pst.1s ‘one, look that (=what) I found’.

iii. J:

mí(r)a que’ cont(r)é, velo look-imp rel find-pst.1s, see-imp=o.3s ‘look that (=what) I found, see it’.

iv. J:

mí(r)a la ata que’ cont(r)é look-imp the tire rel find-pst.1s ‘look the tire that I found’.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

v. J:

mí(r)a- que me quedó look-imp rel io.1s=result-pst.1s ‘look that (=how ) it resulted to me’.

(5) Elia (3;04)

i. E:

esta es la casa que estaba this be.pr.3s the house rel be-impf.3s ‘this is the house that was’.

ii. E:

esta es mi diadema que me trajo una amiga this be.prs.3s my diadem rel io.1s=bring-pst.3s a friend ‘this is my diadem that a friend brough me’.

iii. E: tu pájaro es, las cosas horribles que your bird be.pr.3s the things horrible rel

no me gustan neg io.1s=please-pr.3p

‘your bird is, the horrible things that do not please me’.

iv. E: no te enseño éste, mi premio que me regaló neg io.2s=show-pr.1s this, my price rel io.1s=give-pst.3s

v. E: M: E:

Tana Tana ‘I won’t show you this one, my prize that Tana gave to me’ muerde ‘it bites’ ¿a quién? whom? a unas personas que están llorando to some person-p rel be-pr.3p cry-ger ‘to some persons that are crying’.

3.1.1 Individual preferences In sum, every child shows a different profile of earlier RCs: Flor and Elia prefer lexical heads + RC. Julio selects no overt head with a dominant verb frame: mira ‘look’. Elia adopts a different constructional frame: éste es ‘this is NP rel, and also appends RCs to syntactically free NPs. Flor clearly prefers RCs with free NPs, and only shows a single frame with mira ‘look’ (see Table 2 as a summary). These differences among the early exemplars that children select, are only to be expected from probabilistic recurrent encounters with family usage. Such data confirm that as for syntactic complexity, there is no initial less-complex default and no unique entry point into RC development, but rather a number of individual preferences,

 Cecilia Rojas-Nieto

Table 2. First five RCs exemplars for every child. Predicate framing Lex-fixed

Type of head

Free NP Heads

NP

Pronoun

Det

flor

mira det rel

1

4

4

0

1

elia julio

este es det n rel mira rel

3 4

2 1

5 1

0 0

0 0

lexically driven and partially ritualized, which cut across the criteria that presume to define a possible syntactic default.

3.1.2 Emergent regularities? The earlier RCs, which reveal considerable individual differences and no default, lead us to explore whether we could find, later on, some syntactic factors that affect children’s RCs production, and thus look for possible structural patterns, which may emerge in the course of development. In order to test RC development, and the emergent regularities they might present, we will explore various syntactic variables that would define RCs complexity (Givón 1990): • RCs external syntax: The syntactic freedom or integration of the REL-clause within its main-clause matrix. • RC internal syntax: The adjustment of RC structures vis-a-vis the canonical main clause, in terms of relative pronouns (rel) and their properties, pronominal status and syntactic function; the presence or absence of a ‘syntactic gap’. In every case, we will try to determine whether or not possible regularities have a lexical foundation. We will pay attention to frequency and, if relevant, to developmental chronology.

3.2 External syntax

3.2.1 Freedom and embedding Across the observed period and unexpected from any purely syntactic perspective, most RCs produced by our children do not qualify as embedded clauses (N 170 = 55%). Such RC appear as isolated fragments with no overt syntactic dependence except relfronting and sequencing. RCs are produced under an independent intonation unit (6i) and they may even occur in a different conversational turn, one turn away from their possible head (6ii–6ii).

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

(6) i. Julio (28) painting with watercolour markers which he has been asked to cover J: ayer yo tapé los otos, yesterday I cover-pst.1s det other-p ‘yesterday I put the lid to other ones’

æ que me compó mi papá *det rel io.1s=buy-pst.3s my Dad ‘the ones that my Dad bought for me’ ayer, que me copó yesterday rel io.1s=buy-pst.3s ‘yesterday, that he bought for me’.

ii. Mother and Flor (31), engaged in book reading F: esa es la ropa this be.pr.3s det clothing ‘these is the clothing’ M: ¿cuál? ‘which one’ F: que… de ese nene es rel, of that baby be.pr.3s ‘that, belongs to that baby’. iii. Flor (41) is asking the mother for a particular toy F: juete nenes toy babies ‘babies’ toy’ M: ¿cuál juguete de los niños?, ¿cuál? ‘which children’s toy? which one?’ F: que se mueve rel rfl.3s=move-pr.3s ‘that it moves’. This lack of integration of RC in any larger syntactic frame — even a head, as in the previous exemplars — may also involve Head-RC constructions, not just the RCs alone. In these cases both head and RC are kept apart from any main clause. This is a normal and frequent case in conversation, where the Head-RC construction is given in answer to a recurrent identification question: ¿cuál? ‘which one?’ (7i–ii). (7) i. Julio (26) has demanded of the aunt (A) some water A: a ver enséñame, ¿cuáles vasos? ‘see, show me, which glasses?’ J: (l)os vasos que son (r)icos. det glasses rel be.pr.3p tasty ‘the glasses that are tasty’.

 Cecilia Rojas-Nieto

ii. Elia (25) wants a toy from the upper shelf, which Observer (O) tries to reach. E: la sirena ‘the mermaid’ O: ay! no alcanzo, ¿cuál sirena? ‘auch!, I can’t reach up, what mermaid?’ E: la de… la que tiene… este … cola. det of… det rel have-pr.3s umh tail ‘the one with… the one that has, uh… tail’ O: la que tiene cola, ¡ah! ‘the one that has tail, oh!’ E: ésa. ‘that one’. These and similar cases of Head-RC with no syntactic integration can be argued to be a conversational phenomenon, prompted by the question asking for a referent identification: ¿cuál? ‘which one?’ Such questions establish a context for a free NP-RC to occur in an independent conversational turn (Brucart, 1999). But free Head-RC constructions may also occur inside the child’s own discourse, not only across conversational turns. Children built by themselves free Head-RCs, with no conversational support, when involved in description activities (8i–ii) or when quarrelling or negotiating on reference (8iii), in a sort of ‘referential competition’ situation (Givón 2008). (8) i. Elia (27), asking for a boy she had seen at C’s home E: ¿y… y el hijo que tenías? and and the son rel have-impf.2s? ‘and what about the son you had?’ ii. Flor (37) in a book looking activity, considering an image. F: un sodo que (es)tá mu(y) bonito a fox rel be-pr.3s very nice ‘a fox that is very nice’. iii. Flor (35) discusses with Mom about the brush to be used for combing. F: no es mi cepillo ‘it’s not my brush’ ete mi, et’e mi cepillo, es mi ce-… ete es mi cepillo ‘this is my, this is my brush is my br-… this is my brush’ M: ¿me dejas peinarte? ‘will you let me to comb you?’ F: este que no es ese cepillo this rel neg be-pr.3s that brush ‘this (one) that is not that brush’.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

Various discourse situations beg for elaborated reference. They are the ecological niche of noun phrases with an RC expansion, which add to bare heads a specific information that characterizes or helps identify what the head refers to. These complex noun phrases need not be integrated to any predicate to accomplish their task. Keeping in mind that free RCs or Head-RCs are not exclusively the province of dialogical joint construction, let’s consider more elaborated sequences, where children keep both the lexical head and the RC together, but with an intonation break separating them from their presumed main clause (9i–v). Among these, the clearer cases have both an intonation break and a functional closure (cl): a pronoun or a noun phrase which takes, in the main clause, the position which otherwise the Head-RC might have taken. Then, after the intonation break, the phrase functioning as cl is reformulated and elaborated by means of the Head-RC, as an appositive clarification (9i–iii) (see similar cases in (1iii) and (5iv). These examples all suggest that the Head-RC construction is an independent clause, not integrated into any main clause, perhaps an ‘antitopic’ device. Months later, such a construction will occur as preposed, with similar marking conditions (i.e., a pause and a closure) in a topic position (9iv–9v). (9) i. Elia (23) E: dame ese, ese cuado, ese que give-imp=io.1s that.cl that square that rel

lo tengo, que lo tengues o.3s=have-pr.1s rel o.3s=have-pr.2s

‘give me that one, that square, that one that I have, that you have’.

ii. Flor (34) with Granny (G) G: voy a sacar una ropa que dejé en la lavadora ‘I’m going to take out some clothes that I left in the washing machine’ F: me quiedo taela (=traerla) io.1s=want-pr.1s bring-inf=o.3fs.cl ‘I want to bring them’ la la la opa que que dejates en la bebaloda. the the the clothes rel rel leave-pst.2s in the washing machine ‘the clothes that you left in the washing machine’. iii. Julio (25) and his aunt (A) are looking at some images J: aquí, la eñata (=piñata) (pointing to a stick) ‘here, the piñata’ A: ¿qué? ‘what?’ J: ía, eso, lo que tiene. look-imp that det rel have-pr.3s ‘look, that, what (lit. the that) he holds’

 Cecilia Rojas-Nieto

A: un palo. ‘a stick’. iv. Flor (52) la que me la dio mamá, ahora det rel io.1s= o.3s= give.pst.3s Mum now

la voy a poner o3s=go-pr.1s to put-inf

‘the one Mum gave me, now I will put it…’

v. Flor (64) talking about putting music in the tape-recorder una que tú no te sabes, pónmela one rel you neg rfl.2s= know-pr.2s put-imp=o.3s ‘one you don’t know, put it for me’. To consider the overall presence of free [Head-RCs] in children’s data, see in table 3 the relative proportion of free constructions versus [Head RC] integrated to a predicative frame. Table 3. Syntactic integration of [Head RCs] constructions. Integrated RCs flor elia julio Group

Non-Integrated RCs

Total RCs

Tokens

%

Tokens

%

∑ Tokens

81

.431

107

.569

188

27

.338

53

.662

80

24 132

.545 .423

20 180

.455 .577

44 312

Table 3 shows that more than half of our RCs (=.577) do not fit the common criteria of syntactic complexity–notably embedding in a main clause. And this is the normal situation but for Julio, who as we have already seen, slightly prefers from his earlier records RCs integrated to a particular clausal frame with mira ‘look’. The rest of the children produce more Head-RCs without any external syntactic function; they are not inserted in any clause, and are not syntactically dependent on any predicate. No embedding of one clause inside another is seen here, and in the extreme cases isolated RCs in an absolute position. No functional parallelism is seen here between head and rel-pronoun, (rel), since these heads bear no syntactic relation. What we have instead is a set of referential or deictic forms (pronominal, nominal, det) with an appended RC, which jointly constitute a basic construction. Head and RC have a topic-comment-like relation, whereby the RC is an expansion of the NP rather than an embedded clause. Moreover, early RCs are not necessarily appended to NP or PRO

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

in the same turn. In fact, RCs themselves and [det RC] or [NP RC] frames have in early child language a freedom that RCs rarely keep in adult language (Brucart 1999). In early child language, RCs seem to be parsed as possible independent constructions that may occur alone, or associated to a NP, with which the stable frame of Head-RC is established early on. Consider now table 4, which shows how often free Head-RCs are prompted by conversation, against the frequency of free constructions that rely on children’s own adaptations: when producing an absolute NP-RC in one turn, or when an intonation break or a syntactic closure separates the NP-RC from a possible main clause. Table 4. Free Head RCs conditions. Tokens

%

Dialogic niche

48

.267

Isolated Head-RCs

41

.223

Intonation break

32

.178

59 180

.328 1.000

Lexical or Pronominal closure Total

Here we can attest that important as conversational support is in early child language, dialogue is just one type of context. Though an early one, it only accounts for about a quarter of free Head-RC constructions (.267). Children can and do produce isolated RCs, and independent Head-RCs by themselves, with an intonation break or a closure, pronominal or lexical, an important and frequent preamble to integration (.328). In fact, jointly considered, children’s own free Head RCs with no dialogic support are the most frequent and accessible way for children to add RCs into their discourse. We have again positive evidence that RCs are first produced as isolated pieces, linked with more or less fluency to a noun to form in Head-RCs frames. These are early, self-contained and independent constructions, which serve as a basic niche for children to adopt the use of RCs. But the link between Head and RC may initially be loose, and the Head and RC may be separated by an intonation break or even occupy different conversational turns. These data highlight an important fact noted by Givón (2008): The emergence of RCs in children’s use is to be understood in the functional context of reference tracking. Head-RC constructions serve reference building, which requires no mainclause frame to be accomplished. So, isolated RC or absolute Head-RCs are doing this communicative task in early child language, inserted in unfolding conversation or as a piece in the still fragmentary and unintegrated sequences of child speech. In such a context, the component parts of the complex construction to be — head and matrixclause — are frequently kept apart from the unintegrated, paratactic, ‘adjunct’ RCs.

 Cecilia Rojas-Nieto

3.2.2 Construction frames Notwithstanding that isolated RCs and independent Head-RCs tend to predominate in children’s early production (table 3), we know that from the very start children can insert Head-RCs in various clausal frames, which in the earliest data (table 2) are item-based (mainly, mira ‘look’ and este es ‘this is’). As previous work has argued for English acquisition data (Diessel 2004), we can expect these particular frames will in time increase their frequency and become ubiquitous. The analysis of the lexical frames preferred by children across the period under study proves that indeed the first clausal frames to be adopted — mira ‘look’, and este es ‘this is’ — become the main discursive niches where child RCs occur. But children’s individual preferences emerge early and persist across the period. And our data attest to a wide clausal-frame diversity, more than previous studies would lead us to expect (see table 5 below). Julio, who has shown from his first RCs a preference for the frame mira ‘look’, continues to use this frame as his dominant one (12/24=50%). But he also has a set of secondary frames: hay-había ‘there is/was […]’, este es ‘this is […]’, quiero ‘I want […]’, and

Table 5. Verb frames for [Head-RC] insertion. Frame diversity/ Tokens (39/132)

julio

elia

flor

9/24

12/27

32/81

3

14

14

Lexical Frames: este es [ ] ‘this is’ mira [ ] ‘look-imp’

12

7

hay/había [ ] ‘there is/was’

3

3

5

tengo [ ] ‘I have’

1

1

5

[det rel vb] es FN ‘be-cleft’

6

‘era una vez’ [ ] ‘there was once’

5

dame [ ] ‘give me’ viste/ves [ ] ‘have you seen’

1 1

4 3

llevas [ ] ‘you take’

4

oye [ ] ‘listen-imp’

3

hace ‘he makes’

3

quiero [ ] ‘I want’

1

2

sabes [ ] ’you know’ Various (1 token/each)

2 3

8

18

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

a handful of constructions with various verbs, none of them particularly prominent: ver ‘see’, tapar ‘cover’, prestar ‘lend’, prender ‘turn on’ (see 10i–10iii). (10) i.

Julio (25), ritually starting to tell the Little Red Riding Hood story bía-una vez una capeuceta… que se llama … Ju… Juya. be-impf.3s one time a little-hood, rel rfl.3s=call-pr.3s Ju… Julia ‘Once upon a time there was a Little Red Hood… whose name was…Julia’.

ii.

Julio (26) telling a riddle ete’s una señora que se va llevar uos huevos this is a lady rel rfl.3s=go.pr.3s take.away-inf det eggs ‘there was a lady that was going to take away some eggs’.

iii.

Julio (26), playing with a lamp pende la luz que sí es turn.on-imp the light rel aff be.pr.3s ‘turn on the light that it’s the right one’.

Elia continues to select as her main frame the first one she uses: the identificational presentative éste es […] ‘this is …’ (14/27=52%); but she also adds other presentative constructions: aquí hay […] ‘there is …’ and tengo […], ‘I/we have …’. Together again with a handful of activity verbs: enseñar ‘show’, dar ‘give’, buscar ‘look for’, quitar ‘take out’ (with one occurrence per verb type) (11i–11iii). (11) i.

Elia (23) es unos juguetes que son para jugar be.pr.3s det.p toy.s rel be.pr.3p to play-inf ‘(those) are some toys which are for playing’.

ii.

Elia (24) (d)áme la mochila que tiene … give.imp=io.1s the backpack rel have-pr.3s ‘give me, give me the backpack containing…’

iii. Elia (25) voy a buscar unos zapatos que son así, go.pr.1s to look.for-inf det shoes rel be.pr.3s like.this

mira, éstos look these

‘I’m going to look for some shoes which are like this, look, these ones’.

Flor, with a wider RC production range and no clear dominant frame, adopts the identificational frame éste es […] ‘this is…’ as her preferred one (14/81=.173), together with other frames with ser ‘to be’: the ritual era una vez ‘once upon a time’ and a cleft type

 Cecilia Rojas-Nieto

(lo que pasa/quiero es ‘what I want/it happens is’), though none of them was present among her earliest productions. The preferred frame (éste es ‘identificational’) emerges at (F42: 2;06,4), two months after Flor’s initial verb framed RCs: mira ‘look’, which continues to be present. Besides, Flor uses two more frames built around tengo […] ‘I have … ’ and hay […] ‘there is/are’ which become partly prominent. She incorporates gradually a considerable number of other lexically free and rather diverse verb frames: dar ‘give’, querer ‘want’, poner ‘put’, sacar ‘take out’, etc. (12i–iv). (12) i.

Flor (34) from a window sees somebody has entered the courtyard hay una sudadera azul que entró a la casa there.is a t-shirt blue rel enter-pst.3s to the house ‘there is (a person with) a blue t-shirt coming into the house’.

ii.

Flor (34) asking for some toys in a bird’s cage ¿me das os nenes que tene? io.1s=give-pr.2s det babies rel have-pr.3s ‘will you give me the babies that (the cage) has inside?

iii.

Flor (37) while eating and mentioning dirt things yo ten(g)o p(l)ato que no tiene … . mugue I have-prs.1s dish rel neg have-pst.3s dirt ‘I have a dish that does not have dirt’.

iv. Flor (44) ¿me cuentas un cuento que traiga un libro? oi.1s=narrate-pr.2s a story rel bring-sbj.pr.3s a book? ‘will you tell me a story that is in a book?’ All these individual differences and preferences do not exclude emergent regularities. But in no way do they point towards a unique presentative construction in early RCs. A set of four verb frames, which most children employ, account for half of the embedded RCs tokens (mira ‘look’, éste es ‘this is, hay ‘there is’, and yo tengo ‘I have’). A rich verb diversity is only present in Flor’s data; Julio and Elia have only started their way to develop comparable diversity. In adition, we have to consider that besides the particular verb frames included in table 5, which tend to present an Object (O) slot for Head-RC insertion, the set of verbs jointly presented under various are mainly transitive activity verbs which have also an O as an elaboration site for a Head-RC to occur (buscar ‘look for’, cantar ‘sing’, cazar ‘hunt’, comer ‘eat’, contar ‘narrate’, dejar ‘leave’, enseñar ‘show’, escarbar ‘dig’, llevar ‘take’, necesitar ‘need’, pasar ‘pass’, pegar ‘hit’, prender ‘turn on’, prestar ‘lend’, quitar ‘take away’, sacar ‘take out’, tapar ‘cover’, tirar ‘throw away’). Only a few and infrequent intransitive verbs (estar ‘be.loc’, llamarse ‘be called’, ir ‘go’, llegar ‘arrive’) offer a Subject position as a Head-RC elaboration site. Our data lead us to conclude that there is no specific construction that serves as general default to ‘absorb’ a Head-RC structure. Although some constructions are first used and tend to be preferred as sort of attractor-frames, every child may select

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

particular constructions, which may be similar to the ones other children prefer. But every child’s selection of a particular frame has to be determined in close inspection of the child’s own data. Despite this frame diversity it is not out of question that there may be a reason for the confluence and/or, conversly, the individual preferences we notice. A possible functional assembly may emerge, and may join together a set of frames, which in some not overtly-marked way may have the same effect. In fact, it seems to be the case that what all the observed clausal frames have in common is a slot position […] in the focus domain, where a NP can be inserted. This position tends to be the Object for most attested frames (mira ‘look’, tengo ‘I have’, quiero ‘I want’, etc.); but it may also be a Predicate position, if a particular child adopts éste es ‘this is’ — identificational-equative construction focusing a descriptive NP — as her preferred frame. And it could equally have been an S, if the locative presentative frame aquí está/n — lit. ‘here be.loc-pr.3s/p’, meaning ‘here you are’, ‘here it is/they are’ — had been preferred by some child. This is not the case in these data, but could still be possible for another child. In fact most cases where the slot position is an S involve various intransitive verb frames, well known for introducing topical subjects. Our data do not point towards a unique syntactic position, nor a unique itembased frame, but towards a set of constructions with an elaboration site: a slot where a prominent referring or descriptive NP is inserted; objects are a well known position for new topical information, as are the subjects of intransitive verbs (Clancy 2003; Du Bois 2003; Givón 1984). And the construction este es un/el ‘this is a/the…’ confers prominence on the noun phrase it takes as predicate. What we find across the various syntactic frames where Head-CRs occur in discourse is a NP position inside the focus domain that functions as an elaboration site. And the same informative position may be credited to Head-CRs even when they occur in isolation in ‘absolute’ constructions: both when prompted by WH-questions, which build a focus position for their answers, and as absolute NPs, which are by themselves focal (Zubizarreta 1999).3 As for the various constructional frames involved in this sort of focus elaboration, their commonality could be highlighted by considering them presentative devices (f.i. Alfonso & Melis 2007). I myself prefer the focus interpretation, since dialogue-prompted cases, or absolute Head-RC frames give the same focusing results that conventional presentative frames do, despite not being inserted in any overt ‘presentative’ frame.

3. This elaboration site, which I consider to be in the focus or focal domain, is not meant to be the focus in Lambrecht’s terms. Rather, it prefigures what would be this position in the information structure, itself an expected child’s developemental achievement.

 Cecilia Rojas-Nieto

When one considers this elaboration site in syntactic terms, it is clear that the object (obj) and predicate (pred-nom) are the preferred syntactic positions. They jointly represent 86% of all Head-RCs inserted in a predicate frame, with a frequency range of ca. 84–96% in our children’s data (see Table 6 below for their independent proportions). However, this preference does not seem to depend upon some abstract properties of objects or nominal predicates. We can trace back the dominance of a particular syntactic position to an emergent effect of the lexical predicates in children’s preferred constructions (Clancy 2003; Du Bois 2003), which include a slot that later becomes an elaboration site. Table 6. External position of [Head -RCs]. Syntactic slot of Head RC in a Predicative frame O

%

S

%

PN

%

Obl

%

∑ Embedded RCs

Flor

43

.531

18

.222

15

.185

5

.027

81

Elia

9

.333

2

.074

14

.519

2

.074

27

20 72

.833 .545

0 15

0 .113

3 41

.125 .311

1 8

.042 .061

24 132

Julio Group

In the end, we can make the generalization that it is not the deterministic effect of presentative frames adoption, nor any syntactic variable which define children’s particular sites for [Head RC], but the joint effect of child’s adoption of selective constructions and the informational properties of those constructions that determine the choice of a focus position for the elaboration of reference: either in initial referent presentation or in referential negotiation, the RC serves to elaborate a referential or descriptive noun phrase.

3.3 Internal syntax 3.3.1 Clause adaptations The syntactic complexity of canonical RCs may be defined in terms of the internal adaptations RCs vis-a-vis main-clause structure: the presence of a syntactic gap, the use of a resumptive pronoun, and the dual binding relation of relative pronouns operate: outward in reference to the head NP and inward in reference to the syntactic gap in the RC. From a broader perspective, the internal structural adjustment of dependent clauses is part an parcel of marking them as dependent. Such structural adjustments may range from null in highly-finite languages to maximal adaptation in highly nominalizing languages (Givón, in this volume) and may vary according to

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

dependent-clause type. They involve, most commonly the presence vs. absence of morphological boundary markers, illocutionary force and tense-aspect integration, informational structure restrictions, and argument sharing or integration. At the extreme nominalizing end of this continuum, reduction and reduced finite marking are observed: uninflected or subjunctive-marked verbs, argument sharing reductions or modified case-marking, etc. (Aissen 2004; Givón 1980; Givón, in this volume; Lehman 1988; Van Valin & La Polla 1997). The question to be explored next is to what extent do children’s RCs display any structural adaptation that marks them to be dependent. For RCs, the expected internal adaptations are mainly related to the properties of the rel pronoun: coreference and syntactic rel function. The rel pronoun marks a coreferential link between the head and the internal constituent whose syntactic position rel presumably occupies. The accessibility hierarchy (Keenan & Comrie 1977), considered from a Usage-Based perspective, predicts that the subject rel pronoun will be the most frequent, followed by Object, Indirect Object, and Oblique rels. This point is a little bit tricky in Spanish, a subject-dropping language, since any subjectless RC need not be considered to have the subject represented by the rel pronoun. The very same subjectless clause may occur outside any dependent clause context and independently of the presence of a rel pronoun. Suspending for the moment the presumption that internal subject omission in RCs is a proof of the binding and syntactic properties of the rel pronoun in early child language, we will examine the internal adaptations in RC structure, to see if they appear dependent. As it turns out, RCs in child language are more similar to independent main clauses, despite the presence of a rel pronoun. Thus, some recent experimental work on children’s RCs has shown the impact of such similarity to independent clauses on the acquisition processes (Diessel & Tomasello, 2005). But previous studies of spontaneous data have not emphasized this type of comparison. Given that rel function and binding properties are the most conspicuous aspect of RC complexity, we will consider next the syntactic properties of rel pronouns, in order to evaluate how adapted children’s RCs are to their dependent status, as compared to the independent clause.

3.3.2 rel function in question 3.3.2.1 Thematic association Much like the embeddedness problem, the analysis of rel pronoun status and function in children’s RCs yields an unexpected result: rel pronouns do not necessarily have any syntactic function to fill. RCs may have no clear syntactic gap nor constituent omission whose function the rel pronoun may take. This is clearly seen in cases like (13i–13iii).

 Cecilia Rojas-Nieto

(13) i.

Julio (28) is telling Goldilocks and the three bears story había tles ositos que se cai una exist-impf.3s three bear-dim-p rel rfl.3s=fall-pr.3s one

niña la silla girl the chair

‘there were three bears that a child falls down from the chair’.

ii. Elia (24) me voy a sabe(r) una canción que una casita io.1s=go-pr.1s to know-inf a song rel a house-dim

es bonita be-pr.3s nice

‘I will learn a song that a little house is nice’.

iii. Flor (67) telling a story era una niña que el abuelito se metió a be.impf.3s a girl rel the grandfather rfl.3s=go-pst.3s to

un abujero de ratones a hole of mice

‘there was a girl that her grandfather went into a mice hole’.

Head NPs are expanded in these cases by means of a RC configuration which has no internal gap, nor trace, nor any evidence of structural dependence: only discourse continuity is kept in a lax way and the relation between Head and RC is supported by purely thematic means. These seemingly anomalous RCs, which do not conform to the presumed adult norm, allow us to envisage the way by which children begin to build their early CRs. Children may work on the linking side of these constructions on the basis of a thematic association — a well known procedure to keep discourse continuity — with no syntactic means to bind the RC to its external context and no modification of its internal form. The rel pronoun in such cases is just a local continuity mark.

3.3.2.2 Resumptive linking In addition to thematically-linked RCs that lack internal structural adaptation, we also find RCs with the Resumptive-PRO strategy (res). In such RCs, a pronoun takes the syntactic position expected to be associated with the rel pronoun, leaving the RC with no internal gap. Children thus have a Possessive res in RCs instead of a genitive rel (14i); an io-res in (14ii), instead of a case marked io-rel. (14) i.

Elia (29) is narrating the life of a national hero when he was a child. un pobre que se murió sus papás a poor rel rfl.3s=die-pst.3s pos.3p parents ‘a poor (child) that his parents got dead’ (=whose parents died).

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

ii.

Flor (44) la nena que, la que le pusiste la piyama the baby rel, det rel io.3s=put.on-pst.2s the pajamas ‘the girl that, the one you put her the pajamas’.

In colloquial adult usage, the res-strategy is normal for io-rels (as in 14ii) and is quite frequent for gen-rels (as in 14i), though grammarians proscribe them in formal writing. But children also produce Object-res (15i–iii) and Subject-res RCs (16i–ii), which in adult usage are almost absent (o-res), or plainly ruled out (s-res). (15) i.

Elia (29) asking for some nasal drops (Object res) unas, unas que las tengo aquí some, some rel o.3p=have-pr.1s here some, some that I have them here’.

ii.

Flor (52) (Object res) la que me la dio mamá, la voy a poner det rel oi.1s=o.3s= give-pst.3s Mum o.3s=go.pr.1s to put-inf ‘the one that my Mum gave it to me, I’m going to put it on’.

iii.

Julio (29), excited tells about a baloon his father just bought (Object res) mío, mi (gl)obo que me lo comp(r)ó mine, my balloon rel oi.1s=o.3s=buy-prt.3s ‘mine, my balloon that he bought it for me’.

(16) i.

Julio (30) looking a book (Subject res) te voy a enseñar uno que ése es un caballo io.2s=go.pr.1s to show.inf one rel that be.pr.3s a horse ‘I will show you one which that.one is a horse’.

ii.

Flor (63) (Subject res) es una niña que ella se dumía be.pr.3s a girl that she rfl.3s=sleep.impf.3s ‘(this) is a girl that she got slept’.

What needs to be emphasized here is that in both cases, thematic (THM) and res conditions, RCs do not have an empty place for the rel pronoun. In the res strategy, the internal function is filled by a pronoun (res). Though RCs with res pronouns are considered to be the result of a secondary relativization strategy, and thus may represent a more elaborated types of RCs than simple rel constructions (Comrie & Kuteva 2007), in fact they are structurally simpler. Even if we have an internal link in RCs with a res, these RCs are structurally identical to the independent clause. The rel pronoun is but the linking marker of thematic continuity. RCs with res pronouns thus mark topic continuity procedure, which as an anaphoric operation applies all over the

 Cecilia Rojas-Nieto

grammar, and by itself does not imply structural dependency (Givón ed. 1983, 1984, 1990). We thus propose that res-pronoun RCs be considered an instance of ‘starting small’: formally, there is just a local relation between a head-NP and a RC, marked by a rel pronoun as a topic continuity marker. Functionally, there is a sequence of theme/ topic continuity constructions (Kuno 1987). One may also interpret the res-pronoun strategy as a Usage-Based effect of the adult input. It is indeed possible in adult Spanish usage to have a res pronoun inside the RCs, down from io relativization position and in gen-rel, a usage that can easily become generalized. But for formal written texts, the res-strategy has practically replaced in adult Spanish the marked Genitive-rel cuyo ‘whose’, with a rel + internal possessive (=que su ‘rel pos’). The res-strategy also emerges in various peripheral syntactic positions, but it is infrequent and restricted — near anomalous — with O-rel constructions, and not at all permitted in adult S-rel (Lope Blanch 1986; Palacios 1983). Our children, howerver, exhibit the res strategy even in the S and O positions. So one must recognize the child’s way of building early relative clauses: Using a main clause structure, keeping topic continuity by means of the res pronoun, and adopting the rel pronoun as a surface marker, with no other syntactic integration at all. Such RCs, all by themselves, may be produced as independent clauses; and the rel pronoun then serves as a topic continuity marker.4

3.3.3 Hanging relatives Various other constructions are initiated by the child to function early on as RCs, and are then discontinued as the production of more standard rel clauses kicks in. Thus, the rel pronoun may have no internal function, since the RC is not actually produced, but only announced (17i–17iii). (17) i.

Elia (23) tenemos muchas cosas que… have-pst.1p many things rel ‘we have many things that…’

ii.

Flor (61), doing some gardening tú escarbas, las plantas que, que … you dig-pst-2s, the plants rel, rel ‘You dig the plants that, that…’

4. In our children’s data, Case + rel is just starting to emerge, but we do not have any Case + rel combined with res. This fact may be taken as secondary evidence that rel does not fill any syntactic position in RCs when it co-occurs with res.

iii.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

Julio (29) mira, aquí hay un cuento que… look-imp, here be-pst-3s a story rel ‘look, here it is a story that…’

These fragmented productions make it clear that a rel pronoun is produced after a nominal head, without the child’s having planned the internal configuration of the RC. The presence of the rel pronoun can only be explained here if we credit child’s adoption of a frame N+rel, where N rel keeps a sequential relation, not necessarily a pronominal or syntactic one, in a still unplanned RC construction. Children’s RCs may thus be similar to independent clauses, with no internal function for the rel pronoun but just a thematic link or a resumptive pronoun. In Table 7, below, such RCs jointly represent 17% of all RCs, with little cross-child variation: Flor exhibits 18%, Julio 17% and Elia 15% of such types where the rel pronoun is not associated to an internal gap. Table 7. Internal gap, possibly association to rel.* No gap

Possible gap Internal Gap

Unplanned

THM

RES

[S]

[O]

∑ RCs

Flor

9

11

13

103

59

188

Elia

4

5

4

59

9

80

Julio

3

3

2

21

15

44

Group %

16

19

183 = .587

76 = .244

312

19 54/312 = .173

In this table, res cases have been left out from S-rel and O-rel counting.

3.3.4 Topic continuity or subject gaps in RCs For a subject dropping language, the absence of subject NPs in Spanish RCs can not be used to determine a subject gap, since a ‘gap’ may also appear in main clauses. Subject-internal rel constructions (S-rel) have the same type of structure in Spanish as indendent clauses. This is particularly true in the context of situated reference, a normal condition for child speech (Givón 2008), or referential continuity, a discourse condition for RCs. So, even if the RC configuration seems to have a canonical S-rel, we can argue that there is no need to assume that a RC-specific gap is involved. Thus consider: (18) i. Flor

yo tengo un perro en mi casa que, que hace guau guau I have-pr.1s a dog at my home rel, rel make-pr.3s arf arf

‘I’ve got a dog in my house that, that barks arf arf ’.

 Cecilia Rojas-Nieto

ii.

Julio : ésta es a mamá que está enojada this be.pr.3s the Mum rel be.pr.3s angry ‘This one is the mother that is angry’.

iii

Elia este amarillo es el hijo que está con todos sus papás this yellow be.pr.3s the son rel be-pr.3s with all his parents ‘this yellow one is the son that stays with all his parents’.

This is also seen in table 7: Subject internal continuity in children’s RCs is most frequent. This option represents a total of 58% among all RCs. Besides, Elia shows even a more radical preference for S-rels (=.737) than for O-rels (=.113). Flor has lower rates for S-rels (=.545), since she produces substantially more O-rels (=.277), and even a couple or IO-rels (=.011) (See 16ii above). Finally, Julio has the closest relation between S-rel (=.477) and O-rel (=.341), since he has adopted from the first moment some ritual internal predicates in RCs, which ask for an O-rel (encontré ‘I found’, tengo I have’). We can see this uneven distribution, overall favorable to S-rel, more clearly in Figure 1, where the previous cases of clear “no gap condition” have been left out in order to focus on S-rel and O-rel selection. We thus note that S accessibility is higher than O accessibility in RC binding. This is expected in both RCs and main clauses. And since Subject continuity in RCs is not marked, the S-rel pronoun may be added to RCs that keep the form of an independent

120 100 80

S O IO

60 40 20 0

Flor

Elia

Figure 1. Topic continuity in RCs and posible rel function.

Julio

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

clause. Such cases amount to more than half of all RCs. Topic continuity supported by the S-rel pronoun, with no marking inside the RC, makes for simple unmarked RCs, a dominant pattern here. The few cases with a res subject pronoun (like 16i–16ii, above) show even more clearly an independent clause pattern in the RC. Thus, RCs marked internally as dependent involve only the Object internal rel, to be considered next.

3.3.5 Hard cases made easier: Object rels Object-rel constructions are the main candidates to illustrate the complexities attributed to RCs in general (Fernandez-Duque, in this volume). In Spanish, the presence of an object is considered categorical for transitive verbs, but under certain conditions main clause objects may be omitted (Campos, 1986; Fernández Soriano 1999: §19.4.2). Semantic properties of the verb — those accepting a generic reading and low transitivity: tener ‘have’, querer ‘want’ — and the object’s referential properties (indefinite or generic) and informational status (given), are all involved in these competing options. Accordingly, we can launch into an O-rel analysis without the certainty that we are facing an O-gap associated with rel, or a plain O-dropping. Possible O-dropping in main clauses thus complicates the analysis of Object RCs. We will begin by considering the object rel pronoun as an Object topic-continuity construction, before conceding it the complexities of a bona fide RC. This perspective on RCs conforms to recent experimental studies, which test the impact of O-rel properties in young children’s production and comprehension (Diessel & Tomasello, 2005; Kidd, Brandt, Lieven & Tomasello, 2007). Again, implicit regularities and restrictions — functional, semantic, or constructional — as well as modeled usage are expected to make easier and learnable what seem to defy children’s early capacities. O-rels are in general rather infrequent in early child data. Their emergence depends upon the emergence of a set of specific verbs in RCs. In effect, as we have seen, RCs are externally bound to frames with recurrent predicates, into which Head-RCs are inserted (mira ‘look’, este es ‘this is’, etc.). But there are also recurrent predicates in the RCs themselves; and although these may be child-particular, they are integrated by a well-defined set of verbs in a minimal construction that involves 1s and 2s (S or IO) arguments, with a rel referring to a 3s Head. Apparently, RC-internal predicates furnish the grounding relation between a 3s item, referred by the N head, and the speech-act participants in the RC itself. At one extreme, with the highest O-rel comparative frequency, we find Julio’s production. He adopts from the beginning a couple of RC-internal predicates which require an O (encontré ‘I found’, tengo/tienes I have-you have’). In his production, O-rel has a relatively high frequency (n 17/44 = 38.6%).

 Cecilia Rojas-Nieto

(19) Julio O-rel mira la gata que conté mira lo que tenes allí u regalo, que tiene

‘look the cat I found’ ‘look what you have there’ ‘a present he has’.

On the other side, Elia strongly prefers S-rel, and produces the smallest number of O-rel (n 13/80 = 17%). Among her O-rel constructions (20), tener ‘have’ — with a grasping-handling-possessive reading — shows a particular prominence. Significant for Elia are also ‘giving’ and ‘object transfer’ verbs with a 1s as a benefactive IO. (20)

Elia O-rel ese que lo tengo ‘this one I have’. unas que las tengo aquí ‘one I have her’. mi diadema que me trajo una amiga ‘my tiara a friend brought for me’. mi premio que me regaló una compañera ‘my price a mate presented to me’.

As for Flor, she eventually produces a number of O-rel constructions, but she starts relatively late and only slowly adds different O-rel to her exemplars list. This she does first by using first two construction frames (see 21). Again tener ‘have-hold’ is the preferred one, and ‘giving’ and ‘object manipulation’ verbs (‘give, receive, buy, bring, leave, put’) with a 1s benefactive. But the constructional diversity she develops later leaves out of the question of the productivity of her O-rel constructions (see Table 8). (21)

Flor O-rel ¿me das os nenes que tene? la opa que, que dejates en la bebaloda la capeta que yo te taje mis pasas que me compró Inés

‘will you give me the babies she holds?’ ‘the clothes that you left in the laundry’. ‘the folder I brought to you’. ‘the raisins Ines bought for me’.

Once again, our O-rel sample is highly child specific and, in this sense, accidental. Children’s O-rel constructions represent the individual differences that arise from variable attention focus, processing preferences (Bates, Bretherton & Snyder, 1988), and the vagaries of individual experiences. Still, some generalizations can be made. On the verb-selection side, we find recurrent predicates in particular semantic frames: object manipulation, objects transfer and bare-object contact, all used to ground later expansion and all anchored in speech-act participants. This regularity gives verbs such as ‘have/hold’, ‘give’, ‘receive’, ‘find’ and the like the lion’s share of O-rel constructions (see Table 8, below). What we see in Table 8 is a more or less extended verb inventory for every child, on a par with a fairly rich and diverse object selection. Among the three children, only Flor refers to her own perceptions and desires with RC-internal verbs: ‘which I want’, ‘which I saw’. There is a main grounding verb, semantically general and polysemous — tener ‘have’ — as a sort of O-rel attractor. But the verb inventory also includes concrete

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

Table 8. Internal Syntax: Verb types and Object continuity RCs. Verb types/RCs tokens

elia

julio

flor

7/9

8/15

22/59

5

6

12

5

6

12

1

1

10

1

1

4

1

1

5

1

7

Contact-holding verb tengo-tienes ‘I/you have’ Object transfer verbs encontré/perdí ‘I found, lost’ me dieron/compraron/regalaron/trajeron ‘they gave/bought/presented/brought to me’ Object manipulation verbs poner, sacar, llevar, quitar, dejar, pescar ‘put in, put out, take in, take out, leave, catch’ Speech activity verbs decir, cantar , contar, leer (a story) pedir ‘say, sing, tell , read -a story- ask, demand’ Various activity verbs hacer ‘make’ borrar, comer, romper usar ‘erase’, eat’, ‘brake’, ‘use’ Other non activity verbs

2

quiero ‘want’ ver ‘see’ hay ‘exist’

4 3 1

manipulations (put, move, bring), concrete transactions (give, present, buy), and internal experiences (see, want). These verbs seem to represent the natural history of a child’s relation to concrete objects. The semantic-frame motivation for early verb selection in O-rel constructions is further supported when we see some specific pairings among activity verbs and entities referred as Os in the O-rel construction. Although children first — and mostly — use general verb frames for grounding concrete objects in discourse, they later add specific verb for specific items in O-rel constructions: stories are narrated, books tell things; songs are played (lit. poner ‘put’) on a tape-recorder; paintings and designs are done and erased; and many objects are reported as made. Thus consider: (22)

quelo lo que boló Cami oyes la que pusí ola una que no hemos cantado mía lo que dice mamá, mía lo que hizo papá

‘I want the one that Cami erased (a design)’. ‘did you listen the one I put (some music)’. ‘now one that we have not sang (a song)’. ‘look what it says (the book)’. ‘Mom, look what Daddy has made (a design)’.

 Cecilia Rojas-Nieto

A second type of generalization concerns the semantic category of the items that take the O-rel position. They are all concrete objects. This is true not only for Noun-headed RCs, but also for det-rel constructions, that indexically refer to situated items by means of a generic determiner plus a relative: lo que (lit. det rel the that= ‘what’). This is why noun, pronoun and det headed RCs share the same types of internal predicate. (23) i. ii. iii. iv.

lo que tienes det.n rel have-pr.2s ‘what you hold’. mira lo que encontré look-imp det.n rel find-pst.1s ‘look what I found’. et’es lo que quiero this is det.n rel want-pr.1s ‘this is the one I want’ mila lo que me compró mi papá look det:n rel io.1s=buy-pst.3s my Dad ‘look what my Dad bought for me’.

Supporting the acquisition of syntactically complex O-rels, we find concrete objects surrounding the child’s space, and a rich knowledge of general and specific relations and activities in which such objects are involved. All these are at child’s disposal, helping her identify, present, ground and negotiate concrete referents in a discourse context. Even if Object-rel clauses show a syntactic gap in their internal configuration, the O-rel constructions that our children produce are the product of a conspiracy, involving a rich array of presented objects embedded in conventionalized discourse frames, commenting on the object’s nearest vector the (one who holds it), on caused motion (take, put, in-out), and on social transfer (give, receive, etc.). At the same time, O-rel constructions co-occur mostly with with O-res structures. And among the internal predicates of such clauses, tener ‘have’, and querer ‘want’ are mentioned by grammarians to accept O-dropping in adult Spanish. These facts may not be used as part of the argument here, because one would need to know the syntax of child’s Object dropping in main clauses, a topic that, to my knowledge, remains to be explored in Spanish acquisition.

4. Conclusion Looking towards a possible learning explanation of what otherwise would be considered to be unlearnable and unexplained, this paper argues that early relative-clause

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

constructions do not necessarily have the syntactic properties of the adult complex constructions (Givón 2008; in this volume). We have found in the early production data that RCs are mainly embedded; they tend to be adjoined to plain and syntactically independent heads that exhibit no external function for the head. Our data supports Givón’s proposal that such non integrated chunks are later assembled together into complex constructions via condensation. Such complex constructions emerge later. Their paratactic precursors are produced under separate intonation units, among which one may count isolated NP-RCs, hanging relatives and isolated headless RCs and main-clause-like RCs. Despite the great importance of lexically-specific verb frames for Head-RC insertion, no single presentative construction appears to be the starting — or turning — point in the acquisition of embedded RCs. The set of verbal frames that serve as niches for embedding HEAD-RCs are not presentational frames — unless one dilutes the meaning of “presentational” beyond repair. Rather, such frames may be characterized in terms of the open slot they include in their focus domain: an O for transitive verbs, or an S for an intransitive verb frames. Given such frames, children can build an elaborated reference characterization, either to present a new referent or to settle an argument in referential negotiations. We also have found a significant absence of internal function for rel, and clear evidence pointing towards rel not necessarily being a pronominal form. An RC may not have an empty space to be filled by the relative word: either because it is only thematically linked to the head and there is not a coreferential argument to be bound by rel, or because the coreferential argument is overtly exposed in the RC by means of a res pronominal, leaving no need to consider rel as its anaphoric exponent. Subject internal rels are a dubious frame in Spanish, a subject-dropping language. In most cases, early children’s RCs tend to have the same structure as main clauses. This is particularly striking in the case of S-rel, res-pronoun RCs, and THMlinked RCs. In a parallel with the functional motivation for using Head-RCs in referent tracking, reference building, presentation and negotiation, we infer a functional motivation for RC-internal configuration: notably rel-pronoun marking, the no-gap tendency, and preferred lexical verbs. What at first glance seems to be an adult-like structure turns out, under further scrutiny, to be a main-clause insertion in an RC position. Thus, the main functional effect of a rel pronoun is to mark topic continuity. A rel pronoun need not be considered a pronoun when marked by case, which in early child language does not yet occur, but rather emerges many months after early RC configurations first emerge. So, although the main source of rel binding would refer to RC subject, as predicted by the NP hierarchy, this prominence does not guarantee a relativization

 Cecilia Rojas-Nieto

scenario, but only subject-based topic continuity. The cases where the rel-pronoun has a clear internal function are thus limited to Object RCs. We need to know more about O-dropping in early child language as well as in adult usage, O-dropping with tener ‘have’, querer ‘want’, ver ‘see’, and hay ‘there is’ opens the door for extending the topic-continuity interpretation from S-rel-pronouns to O-rel-pronouns. Even if O-rel-pronouns should end up as a clear case of RCs with an internal gap, children’s O-rel clauses may still benefit from simpler processing. The internal structure of the RC may have been modified (although O-rel frames are the main source of res), but the internal configurations of the RCs is easier to resolve because of the use of a handful of semantic frames, whose limited types and conventionality make it possible for the child to apply them routinely and have a high memory. We can conclude that as far as purely syntactic criteria may characterize early RCs as complex frames (embeddedness, rel dual — external-internal — and functional parallelism between head and rel), our children’s data do not support these criteria. There are, of course, many syntactic regularities in the children’s early RCs. But these regularities do not conform to the abstract predictions of complex dependence or functional parallelism. Only the NP accessibility hierarchy, which conforms to topic continuity, seems to be predictive. But even here, we cannot definitively state that rel-prounoun takes the role of an S, nor that the RC manifests a syntactic dependency to the main clause, which may not even exist if the RC — or the NP-RC — is isolated under a separate intonation contour. Hard as it may be to characterize on its own terms the child’s development of complex syntactic structures, our children seem to operate in a way that can be described as ‘starting small’. Children treat RCs as chunks that may be produced alone; they recruit some host frames, not necessarily with a verb, and they may locally affix a RC onto a referential or descriptive NP, without implicating greater complexity. The fact that some predicate frames recur from child to child and emerge as the preferred ones for individual children points towards a preferential syntactic position for the insertion of lexical NPs and their associated RCs. But this is, more likely, an informational effect rather than an abstract, apriori syntactic condition. We have considered all these syntactic twists — having a res pronoun, dispensing with the gap, taking a HEAD + RCs alone outside a host syntactic frame, or selecting some particular syntactic frames — as ‘starting small’ effects. What we seem to observe here is a set of devices that act to reduce the complexity of RCs; something like ‘divide and conquer’ (Elman 1993; 2005), ‘adopt a construction frame’ (Diessel & Tomasello, 2001), ‘use a main clause as an RC’ (Diessel & Tomasello, 2005); which makes it easy to get a rel-clause under the functional impetus of complex reference building.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

List of abbreviations aff cl det dim ger imp impf neg pr pst

Affirmation Closure Determiner Diminutive Gerund Imperative Imperfect Negation Present Past

rc rel rfl res sbj f m n p s

Relative clause Relative pronoun Reflexive Resumptive pronoun Subjunctive Feminine Masculine Neuter Plural Singular

References Aissen, J. 2006. Taller de Complementación (agosto, 2006). San Cristóbal de las Casas, Chiapas: CIESAS-Suroeste. Alfonso, M. & Melis, C. 2007. La alternacia en el orden del sujeto en la oración intransitiva del español. Primer Encuentro de Estudios de Gramática del Español. Mayo 2007. México: El Colegio de México. Barriga, R. 2002. La producción de oraciones de relativo en niños de seis años. In Estudios sobre habla infantil en los años escolares. ‘…un solecito calientote’. México: El Colegio de México. Bates, E., Bretherton, I. & Snyder, L. (Eds). 1988. From First Word to Grammar: Individual Differences and Dissociable Mechanisms. Cambridge: CUP. Bello, A. 1847/1988. Gramática de la lengua castellana destinada al uso de los americanos. Ed. crítica de Ramón Trujilo con las notas de R.J. Cuervo (1988). Madrid: Arco Libros. Bowerman, M. 1979. The acquisition of complex sentences. In Language Acquisition. Studies in First Language Development, P.M. Fletcher & M. Garman (Eds), 285–305. Cambridge: CUP. Braine, M.D.S. 1976. Children’s first words combinations. Monographs of the Society for Research in Child Development 41. Brucart, J.M. 1999. La estructura del sintagma nominal: Las oraciones de relativo. In Gramática descriptiva de la lengua española, Vol. 1, I. Bosque & V. Demonte (coords), 395–522. Madrid: Espasa. Campos, H. 1986. Indefinite object drop. Linguistic Inquiry 17(2): 354–359. Comrie, B. & Kuteva, T. 2007. Complexity and relative clauses in creole languages, Seminario de complejidad sintáctica. Noviembre 2007. Hermosillo: Universidad de Sonora. Clancy, P.M. 2003. The lexicon in interaction. Developmental origins of preferred argument structure. In Preferred Argument Struc00000ture: Grammar as Architecture for Function, J. Du Bois, L. Kumpf & W. Ashby (Eds), 81–108. Amsterdam: John Benjamins. Dabrowska, E. 2001. Learning a morphological system without a default: The Polish genitive. Journal of Child Language 28: 545–574. Dabrowska, E. & Lieven, E. 2005. Towards a lexically specific grammar of children’s question constructions. Cognitive Linguistics 16(3): 437–474.

 Cecilia Rojas-Nieto Diessel, H. 2004. The Acquisition of Complex Sentences in English. Cambridge: CUP. Diessel, H. & Tomasello, M. 2000. The development of relative constructions in early child speech. Cognitive Linguistics 11: 131–152. Diessel, H. & Tomasello, M. 2001. The acquisition of finite complement clauses in English: A usage based approach to the development of grammatical constructions. Cognitive Linguistics 12: 97–141. Diessel, H. & Tomasello. M. 2005. A new look at the acquisition of relative clauses. Language 81(4): 882–906. Du Bois, J. 2003. Argument structure. Grammar in use. In Preferred Argument Structure: Grammar as Architecture for Function, J. Du Bois, L.-E. Kumpf & W.J. Asby (Eds), 11–60. Amsterdam: John Benjamins. Echeverría, M. 1978. Desarrollo de la comprensión infantil de la sintaxis española. Concepción, Chile: Universidad de Concepción. Elman. J.L. 1993. Learning and development in neural networks: The importance of starting small. Cognition 48(1): 71–99. Elman, J.L. 2005. Learning and generalizations. Lesson from neural networks. Plenary Lecture. X International Congress for the study of child language. Freie Universität Berlin–IASChL (26 July, 2005). Berlin. Fernandez-Duque, D. (in this volume). Cognitive and neural underpinnings of syntactic complexity. Fernández Soriano, O. 1999. El pronombre personal. Formas y distribuciones. Pronombres átonos y tónicos. In Gramática descriptiva de la lengua española, Vol. I, I. Bosque & V. Demonte (coords), 1209–1274. Madrid: Espasa. Gathercole, V.M., Sebastián, E. & Soto, P. 1999. The early acquisition of Spanish verb morphology. Accross the board or piece meal knowledge. The International Journal of Bilingualism 3: 183–182. Givón, T. 1980. The binding hierarchy and the typology of complements. Studies in Language 4(3): 333–377. Givón, T. (Ed.) 1983. Topic Continuity in Discourse, TSL #3, Amsterdam: J. Benjamins Givón, T. 1984. Syntax. A Functional-Typological Introduction, Vol. I. Amsterdam: John Benjamins. Givón, T. 1990. Syntax. A Functional-Typological Introduction, Vol. II. Amsterdam: John Benjamins. Givón, T., in this volume. Multiple routes to clause union: The diachrony of syntactic complexity. Givón, T. 2008. The ontogeny of relative clauses. How children learn to negotiate complex reference. In The Genesis of Syntactic Complexity, T. Givón, Ch. 8. Amsterdam: John Benjamins. Hale, K. 1976. The adjoined clause in Australia. In Gramatical Categories in Australian Languages, R.M. Dixon (Ed.), 78–105. Canberra: Australian Institute of Aboriginal Studies. Hurtado, A. 1984. Estructuras tardías en el lenguaje infantil. México: SEP-OEA. Keenan, E. & Comrie, B. 1977. Noun phrase accessibility and universal grammar. Linguistic Inquiry 8: 63–99. Kidd, E., Brandt, S., Lieven, E. & Tomasello, M. 2007. Object relatives made easy: A crosslinguistic comparison of the constraints influencing young children’s processing of relative clauses. Language and Cognitive Processes 22: 860–897. Kuno, S. 1987. Functional Syntax: Anaphora, Discourse and Empathy. Chicago IL: The University of Chicago Press. Lambrecht, K. 1988. There was a farmer had a dog: Syntactic amalgams revisited. Berkeley Linguistic Society 14: 319–339.

‘Starting small’ effects in the acquisition of early relative constructions in Spanish 

Lehman, C. 1988. Towards a typology of clause linkage. In Clause Combining in Grammar and Discourse, J. Haiman & S. Thompson (Eds), 181–225. Amsterdam: John Benjamins. Limber, J. 1973. The genesis of complex sentences. In Cognitive Development and the Acquisition of Language, T. Moore (Ed.). New York NY: Academic Press. Lieven, E., Pine, J. & Baldwin, G. 1997. Lexically-based learning and early grammatical development. Journal of Child Language 24: 187–220. Lope Blanch, J.M. 1986. La despronominalización de los relativos, Estudios de lingüística española, 119–136. México: Instituto de Investigaciones Filológicas-Universidad Nacional Autónoma de México. McClure, C., Pine, J. & Lieven, E. 2006. Investigating the abstractness of children’s early knowledge of argument structure. Journal of Child Language 33(4): 693–720. Moreno Cabrera, J.C. 1999. Las funciones informativas: Las perífrasis de relativo y otras construcciones perifrásticas. In Gramática descriptiva de la lengua española, Vol. III, I. Bosque & V. Demonte (coords), 4245–4302. Madrid: Espasa. Newport, E. 1990. Maturation constraints on language learning. Cognitive Science 14: 11–28. Palacios, M. 1983. Sintaxis de los relativos en el habla culta de la ciudad de México. México: Instituto de Investigaciones Filológicas-Universidad Nacional Autónoma de México. Prideaux, G.D. & Baker, W. 1986. Strategies and Structures: The Processing of Relative Clauses. Amsterdam: John Benjamins. Rojas-Nieto, C. 2003. Early acquisition of verb inflexion in Spanish. A Usage-based account. Psychology of Language and Communication 7(2): 33–56. Rojas-Nieto, C. 2007. La base de datos ETAL: Etapas tempranas en la adquisición del lenguaje. Jornadas Filológicas 2005. México: Instituto de Investigaciones Filológicas-Universidad Nacional Autónoma de México. Rojas-Nieto, C. 2008. Before grammar. Cut and paste in early complex sentences. (submitted). Seidenberg, M.S. & MacDonald, M.C. 1999. A probabilistic constraints approach to language acquisition and processing. Cognitive Science 23: 569–588. Sicuro Corrêa, L. 1995. An alternative assessment of children’s comprehension of relative clauses. Journal of Psycholinguistic Research 24: 225–262. Tavakolian, S. 1981. The conjoined clause analysis of relative clauses. In Language Acquisition and Linguistic Theory, S. Tavakolian, (Ed.), 167–187. Cambridge MA: The MIT Press. Tomasello, M. 2000. Do young children have adult syntactic competence? Cognition 74: 209–253. Tomasello, M. 2003. Constructing a Language. A Usage Based View of Language Acquisition. Cambridge MA: Harvard University Press. Thompson, S. 1998. A discourse approach to the crosslinguistic category of adjective. In Explaining Language Universals, J. Hawkins, (Ed.), 167– 185. London: Basil Blackwell. Van Valin, R.D. & LaPolla R. 1997. Syntax. Cambridge: CUP. Zubizarreta, M.L. 1999. Las funciones informativas: Tema y foco. In Gramática descriptiva de la lengua española, Vol. III, I. Bosque & V. Demonte (coords), 4215–4224. Madrid: Espasa.

The ontogeny of complex verb phrases How children learn to negotiate fact and desire T. GivÓn

University of Oregon This paper investigates the acquisition of V-complement constructions (complex VPs) by English-speaking children ca. age 1;8-to-2;9. It suggests that the child acquires these constructions during intensive epistemic or deontic modal negotiations with the adult. In the earliest stage, the main-plus-complement construction is spread over adjacent child-adult or adult-child conversational turns ( Ochs et al. 1979). The early precursor of the complex VP construction is thus paratactic, with the two clauses falling under separate intonation contours. Only later on is the construction condensed into a complex syntactic construction under a single intonation contour, produced by the child alone. The early use of these constructions is as direct speech acts, be they epistemic or deontic (Diessel 2005), whereby the semantic focus resides in the complement clause, and the main clause acts as a modal operator. But this is true of both the children and their adult interlocutors, and is also characteristic, at the text-frequency level, of adult oral language (Thomson 2001). However, this characterization of complex VPs is semantic rather than syntactic.

1. Overview* The acquisition by children of complex verb-phrases (V-complement constructions) has been studied recently in two ground-breaking works, Diessel & Tomasello (2001) and Diessel (2005). Among their many findings, four closely-related observations concern us most here. They may be summarized as follows: i. In the early stage of child use of V-complement constructions, the main verb acts as a grammaticalized modal operator, epistemic or deontic, on the complement clause. ii. Consequently, V-complement constructions in early childhood are acquired holistically, as single propositions whose semantic focus is on the complement clause. *I am indebted to Holger Diessel for his work, which inspired this study, as well as for generous comments on earlier versions of the manuscript. I am also greatly indebted to Brian MacWhinney for use of the CHILDES data base, as well as for many stimulating comments.

 T. GivÓn

iii. Only later do children develop the use of V-complement constructions as a twoproposition construction, the presumed adult norm, in which each clause carries its own semantic weight. iv. Syntactically too, the child acquires complex VPs initially as simplex single-clause constructions, and only later re-analyzes them — by expansion — as complex twoclause constructions. Diessel and Tomasello’s observation (iv), if it holds, implies that the ontogeny of complex VPs proceeds in the exact opposite direction from their diachrony. That is, the main trend in the diachrony of complex clauses is for a paratactic twoclause configuration, falling under two separate intonation contours, to undergo combination (‘condensation’) and become a complex clause under a single intonation contour (Givón 2008, ch. 4,5). While taking the findings of Diessel and Tomasello as an important point of departure, I will suggest here that there is a way in which the essentially-semantic observations (i) and (ii) above can be granted without necessarily subscribing to the essentially-syntactic (iv). As for observation (iii), its specificity to children — as against adults — bears re-examining. The main thrust of my argument concerning Diessel and Tomasello’s observation (iv) harkens back to a body of work that is by now, deservedly, part of the acquisitional canon; that is, Ervin-Tripp (1970); Scollon (1976) and Ochs et al. (1979). Those works suggest that in the early stages of child communication, between 1–2 years, both propositions (semantics) and clauses (syntax) are distributed over multiple child-adult or adult-child conversational turns. And that such cross-turn shared constructions are the true precursors of the single-turn clauses that emerge in subsequent stages of child communication.1 To quote Ochs et al.’s (1979) conclusions: “… caretaker and child together construct a single proposition. We suggest that a child may learn how to articulate propositions through such a mechanism. That is, she may learn how to encode propositions by participating in a sequence in which she contributes a component of the proposition … We may ask: To what extent is a child able to encode the proposition he wishes to convey in a single utterance? … Which dimension of the utterance context (verbal and/or nonverbal) does the child exploit … ? …” (1979: pp. 267–268)

. The cross-turn construction of clauses, propositions, and discourse topics is just as prevalent in adult communication (Chafe 1994, 1997; Ervin-Tripp & Küntay 1997; Linell & Karolija 1997; inter alia).

The ontogeny of complex verb phrases 

In this paper, I hope to show that cross-turn shared constructions are copiously present in the early stages of acquisition of complex VPs, and should be considered as their developmental precursors. Much of the seeming discrepancy between Diesel and Tomasello’s observations and my own hinges on a simple point of methodology. Diessel and Tomasello tracked down the first appearance in the child’s use of canonic adult-like V-complement constructions, such as: (1) a. DEONTIC: I want to eat the apple. b. DEONTIC: Let me have a toy. c. EPISTEMIC: I know (that) it is broken.

(⊃ please give me an apple) (⊃ please give me a toy) (⊃ it is broken, I’m certain)

They noted that when the child first produces such constructions, their semantic value is simplex, so that (1a) and (1b) are simple, unitary direct manipulative speech-acts of request, and (1c) is a simple, unitary direct declarative speech-act of emphatic assertion. Only later on in development do the semantically-complex usages of such constructions emerge, most conspicuously with 3rd person subjects; as in, respectively: (2) a. DEONTIC: She wanted to eat an apple. b. DEONTIC: He let her have the toy. c. EPISTEMIC: They knew (that) it was broken. In such later uses, both the main and complement propositions presumably have independent semantic values. Note, however, that the developmental transition from the early stage (1) to the later (2) is strictly semantic, and has no discernible syntactic correlates. Diessel and Tomasello’s assertion (iv) thus remains unsupported. But — do the early complex VP constructions in (1) have earlier yet paratactic precursors? What I propose to show here is that cross-turn sharing of complex constructions a la Ervin-Tripp (1970); Scollon (1976) and Ochs et al. (1979) is the real precursor of the early-stage child usage in (1). Such cross-turn sharing of complex structures is richly documented in the data base — the CHILDES transcripts — studied by Diessel (2005). But in order to identify such paratactic precursors, one must look not only at isolated utterances produced by the child, but at larger chunks of multi-turn interactions between the child and adult. The second topic that begs re-examination is Diessel and Tomaesello’s characterization of the adult standard for complex VPs, the benchmark children reach later in development (observation (iii) above). The adult standard — the child’s presumed target construction — is viewed by Diessel and Tomasello as consisting of two semantically independent propositions, thus of two syntactic clauses. Their observations (i), and (ii) thus sets up a stark difference between early child and adult usage. The characterization of adult complex-VPs as consisting of two independent propositions harkens back to Chomsky’s (1957, 1965) early transformational

 T. GivÓn

grammar. This characterization was challenged head-on by Thompson (2002; see also Thompson & Mulac 1991), two studies that have more recently been themselves challenged by Boye & Harder (2007) and Newmeyer (2008). Along lines compatible with Thomson’s work — but only up to a point — I would like to suggest that the adaptive-communicative motivation for recruiting epistemic and deontic main verbs into complex VP constructions is the same in both diachrony and ontogeny — the creation a deontic or epistemic modal envelope on the complement proposition. In diachrony (adult) as in ontogeny (child), the use of such constructions with two independent semantic foci, as in (2) above, is a secondary development. Such usage characterizes neither the early stages of ontogeny and diachrony, nor the bulk of everyday oral — non-academic — adult usage. But as both Boye & Harder (2007) and Newmeyer (2008) note, high frequency does not mean 100%. And when adults talk about the mental states of third persons in the past or future, epistemic and deontic verbs assume an independent semantic value. This argument is indeed complex, involving first the diachronic development of modal expressions (Givón 1979; Hopper & Traugott 1993; Heine et al. 1991; Heine & Kuteva 2007). It also involves the question of what text-type, or rather what communication type, is the prototype for human language. At the very least, I will show that at the text-frequency level, the use of V-complement constructions in oral informal adult communication does not differ significantly from that of early childhood. And thus that non-academic, non-philosophical adult oral usage conforms to Diessel and Tomasello’s semantic characterization of early childhood usage.

2. Data-base The transcripts of child-adult communication studied here were taken from the CHILDES English data-base.2 This is the same data-base studied by Diessel (2005), with two children (Naomi, Nina) also appearing in Diessel’s (2005) 5-children sample, and one child (Eve) added. For each child, three developmental stages were selected by informal criteria. In stage-I, very few examples of complex verb phrases were found. In stage-II, more. In stage-III, many more. Approximately 60 printed pages of the CHILDES transcripts were studied for each child at each stage, aiming for contiguous single recording sessions whenever possible. Since the original transcripts on hard disk are often un-paginated, page numbers for later reference were added after the printing.

. See MacWhinney & Snow (1985). Access to the CHILDES electronic data-base was obtained courtesy of Brian MacWhinney.

The ontogeny of complex verb phrases 

In the case of Naomi’s stages I and II, multiple recording sessions were combined to make up the aimed-for bulk.3

3. Modal interaction units 3.1 Simple modal interactions The use of deontic and epistemic main verbs as modal operators on complement propositions does not occur in a communicative vacuum. To appreciate how such constructions are used, one must study the communicative goal-context within which they are embedded. I will call these contexts in child-adult communication modal interaction units (MIUs). If the CHILDES transcripts are any indication, such units can be broadly classified as carrying either epistemic or deontic goals. By epistemic I mean “pertaining to how a person views the facts of the world, or how they view another person’s view of such facts”. By deontic I mean “pertaining to what one person wants to do, or what they want another person to do for them”. As Diessel (2005) has observed, in the early stages of child language acquisition these modalities are used in a more restricted sense, as direct speech-acts. That is, epistemic: “how I view the facts of the world, or how I view your view of such facts”. And deontic: “what I want to do, or what I want you to do for me”. What I have done with the ca. 60 pp. of CHILDES transcripts for each of the 3 children at each of the 3 developmental stages, is isolate and extract all the MIUs — coherent chunks of diadic communication — that surround each complex VP construction (or a cluster thereof) in the text, be they deontic or epistemic. Some of these MIUs are

. The transcribed recording sessions for each of the three children from the CHILDES database studied here are as follows: EVE: Stage-I: age: 1;9; date: 1–14–63; pp. 1–69; Stage-II: age: 1;10; date: 2–25–63; pp. 1–60; Stage-III: age: 2;0; date: 4–29–63; pp. 1–62; NAOMI: Stage-I: age: 1: 10: 10; date: 4–18–70; pp. 1–2; (Naomi.11) age: 1: 10: 14; date 7–22–70; pp. 27–39; (Naomi.13) age: 1;10;17; date: 2–25–70; pp. 40–53; (Naomi.14) age: 1;10;18; date: 4–26–70; pp. 54–62; (Naomi.15) Stage-II: age: 2;0;02; date: 6–26–70; pp. 1–41; (Naomi.35) age: 2;0;18; date: 6–26–70; pp. 42–62; (Naomi.38) Stage-III: age: 2;2;25; date: 9–08–70; pp. 1–52 NINA: Stage-I: age: 1;11; date: 11–05–70; pp. 1–65; (Nina01.cha) Stage-II: age: 2;3;18; date: 3–07–71; pp. 1–54; (Nina.18) Stage-III: age: 2;9;26); date: 9–15–71; pp., 1–57; (Nina.34)

 T. GivÓn

short and simple, either purely deontic or purely epistemic. Some examples of those are:

(3) Simple modal interactions

a. Deontic: (Eve-I, p. 2) EVE: Napkin. MOT: Oh, do you want a napkin too?

(request) (offer)

b. Deontic: (Eve-I, p. 3) EVE: Fraser blow nose, blow nose. MOT: Wipe your nose? Can you blow?

(request) (offer)

c. Deontic: (Eve-I, pp. 15–16) EVE: Bottle. (request) MOT: What? (request for interpretation) EVE: Eve … (request) MOT: Do you want to taste it? (offer) Let’s see if Sarah would like to have a drink (manipulation) EVE: Eve want some too. (request) d. Epistemic: (Eve-I, p. 57) EVE: Eating bread too. (observation of facts) MOT: She’s eating bread too, I think. (quantification of facts) e. Epistemic: (Eve-I, p. 59) FAT: What are you doing? EVE: Have shower hat. FAT: Well, I know you are wearing a shower hat. EVE: Eve wear-ing shower hat.

(question of facts) (statement of facts) (quantification of facts) (statement of fact)

f. Epistemic: (Eve-I, pp. 55–56) EVE: Got [=dog] bark-ing. (statement of fact) RIC: He got what? (question; mis-communication) EVE: Got bark-ing. [x5] (repeated statement of fact) MOT: There’s a dog barking outside (interpretation of … Yeah. statement) COL: I’m not sure. Yeah, I think it is. (epistemic quantification) I’m sure it is. (epistemic quantification) Instead of saying ‘dog’ she says ‘got’ (resolving mis-communication) EVE: Got eat-ing bread too. (statement of fact)

The ontogeny of complex verb phrases 

With one partial exception (‘Eve want some too’ (3c)), all the complex epistemic and deontic constructions in (3) are contributed by the adult. But whoever the contributor, these complex grammatical constructions are embedded inside a modal interaction context, an envelope within which — and through which — the two participants negotiate their deontic or epistemic goals, or resolve their epistemic or deontic conflicts. It is thus the entire multi-turn MIU that should be counted as the developmental precursor for the child’s eventual acquisition of these complex syntactic structures, and thus of transacting more effectively deontic and epistemic negotiations. Such verbal sophistication is almost entirely absent in our Eve-I (age 1;9), Naomi-I (age 1;10) and Nina-I (age 1;11) transcripts. The collaborative nature of modal interactions is evident in the child’s interspersed contribution, often mere fragments of the intended proposition (see Givón, 2008, ch. 10). Thus in (3c) above, Eve first contributes the object (‘bottle’), then the subject (‘Eve’). Only at the very end, after the mother has interpreted the deontic goal correctly and used the appropriate deontic verb (‘Do you want to taste it?’), does Eve produce a full proposition (‘Eve want some too’), albeit with a simplex use of the modal verb — nominal object rather than verbal complement.

3.2 Complex modal interactions Often, especially in longer MIUs, the modal intent of the negotiation may shift in midstream. The change may involve: •• Who initiates, and thus controls, the interaction. •• Shift(s) of modality in mid-MIU by either interlocutor, from deontic to epistemic or vice versa. The modal complexity of MIUs is more conspicuous in the later, stage-II or stage-III transcripts. Thus, consider (4) below, where the mother, rather characteristically, recruits an epistemic argument, together with its attendant modalgrammatical machinery (here two quotative verbs) to settle the initial deontic conflict (Naomi-III, p. 4): (4) NAO: MOT: NAO: MOT: NAO:

Give me a diaper. Yes, I’ll get you a diaper, honey. You let go again. Okay, want to come down and get this diaper changed? No. You told me about it, Nomi. You said: “Mommy change my diaper”. Boom Mommy.

(request = DEONT) (promise = DEONT) (manipulation = DEONT) (offer = DEONT) (refusal = DEONT) (past-quotative = EPIST) (past-quotative = EPIST) (utter disdain = DEONT)

 T. GivÓn

But the child herself is quite capable of pulling the same trick, indeed of replying in modality-shifting kind, as in (Naomi-III, pp. 51–52): (5) NAO: MOT: NAO:

I want to sit by the tape-recorder. (request = DEONT) I’m sorry, you’re too heavy. (regretted facts = EPIST) and you’re going to break it. (dire prediction = EPIST) Why don’t we do something else? (manipulation = DEONT) It’s not brok-en! (counter statement of facts = EPIST) MOT: Well, you are breaking it now honey. (counter statement of facts = EPIST) You are hurting it. (counter statement of facts = EPIST) Verbally helpless just 4 months earlier, Naomi has learned well, indeed from a master negotiator, the subtle art of modal fencing.4 And the rapiers wielded in service of one’s modal goals — be they deontic or epistemic — is this relatively small group of complement-taking main verbs, the so called modality verbs, manipulation verbs, and

. The following example is taken out of a fictional account of an adult epistemic-modal fencing match between two characters, Momma and Mrs P.J. King (Pearson 1985; modal operators are bold-faced): … “Pepsi Cola” she said. “Yes, I believe is was Pepsi Cola because I’m near certain it was Mr. Womble who ran the Nephi outfit”. And Momma sat straight up and said, “Helen?” … But Mrs. Phillip J. King just went straight on and said, “It had to be Pepsi Cola. He owned the bottling plant you know in Burlington. I mean his daddy, now I don’ think he ever owned it himself, but his daddy did and made a killing putting out Pepsi Cola until he sold the business and made another killing doing that”. Momma said it was just a ton of money that changed hands. “She was brought up in Burlington you know. “But Helen”, said Momma … “And they tell me his wife was just a gorgeous woman but not from around here … Momma said he went out and got one all the way from Delaware or Ohio, she couldn’t ever remember exactly which, but I imagine it was Delaware since P.J. tells me … that Delaware is one of your urban states … and P.J. says there’s plenty of money in Delaware mostly on account of the Duponts, and she might have even been a Dupont herself, anyway I don’t know that she wasn’t and she was probably from Delaware I imagine, which is where they all come from …“ “Wasn’t it cookies instead of Pepsi-Cola?” Momma wanted to know. “Didn’t Mr. Alton’s Daddy make those savannahs with white cream filling and those little oval shortbread cakes that came in the blue sack?” And Mrs. Phillip J. King got a little hot on account of the cream-filled savannahs and the shortbread cakes and she said to Momma, “Now Inez, he might have dabbled in cookies later but I can tell you for a fact it was Pepsi-Cola at the first because Momma said it was Mr. Womble at the Nehi and Mr. Foster at the Coca-Cola and Mr. Tod W. Smith at the Sundrop and Mr. Nance at the Pepsi-Cola, and Momma herself told me it was Pepsi-Cola that made him his money but I don’t ever recall a whisper of cookies passing her lips …” …” (T.R. Pearson, A Short History of a Small Place, 1985, pp. 193–195)

The ontogeny of complex verb phrases 

p erception-cognition-utterance verbs (Givón 2001; ch. 12). Eventually, the mother’s modal fencing skills in (5) are too much for her daughter.

3.3 Boundaries of modal interaction units Sometimes the initial boundary of the MIU is not altogether obvious, especially in cases when a long interaction precedes the complex grammatical-modal form — without the use of any complex grammatical expressions in that preceding sequence. Thus consider (6) below, a lengthy and relatively conflict-free epistemic negotiation (Nina-I, p. 3): (6) NIN: Big. MOT: Yeah. NIN: Big crocodile. MOT: Big crocodile. It sure is.

(statement of fact = EPIST) (agreement on facts = EPIST) (expanded statement of facts = EPIST) (quantification of facts = EPIST)

NIN: Rabbit. (topic shift; statement of new facts = EPIST) NIN: Little rabbit. (expanded statement of new facts = EPIST) MOT: That’s a little rabbit. (agreement on of facts = EPIST) NIN: On a bicycle. (expanded statement of facts = EPIST) MOT: Oh, is the rabbit riding (challenging question of facts = EPIST) on the bicycle? NIN: Yeah. (assent of facts = EPIST) MOT: What is the rabbit (question of facts = EPIST) doing? EVE: Fall down. (statement of facts = EPIST) The second modal interaction in (6), involving the new topic (‘rabbit’) and the complex expression with the progressive auxiliary (‘is’), may easily be detached from the first one (topic = ‘crocodile’), marked with the epistemic quantifier ‘be sure’, without any loss of coherence to either. In deciding the boundaries of MIUs, a cluster of criteria were considered, most salient among them:

•• Economy: Are one or more complex modal-grammatical expressions clustered together naturally?

•• Contextual relevance: Is the immediate context more relevant than the distant one? How immediate is immediate?

•• Thematic coherence: Is the thematic thread maintained or interrupted? When these criteria come into conflict, they must be weighed — and sometimes weighted — against each other. In this study I have elected, whenever possible, to not let modal discontinuity by itself — deontic-to-epistemic or epistemic-to-deontic shifts — be the sole motivation for positing an MIU boundary, as long as the topical

 T. GivÓn

thread is not broken by the modality shift. This choice, I think, is well-supported by what we have noted in (4) and (5) above about modality shifting within an MIU. Conversely, I consider a successful topic shift by either the adult or the child as a good grounds for inserting an MIU boundary. This may be seen in Nina’s abrupt shift in (6) above, from ‘crocodile’ to ‘rabbit’. Considerations of both topical and modal coherence thus form the bulk of the motivation for packaging MIUs the way I did.

3.4 Identifying the child’s speech-act intention In the early stages (I, II), the child’s modal intention is often left unmarked. How does one, working from the CHILDES transcripts, determine the speech-act value of the child’s elliptic utterances?5 The question can be recast as follows: How do the adult interlocutors divine, seemingly without fail, the child’s modal intention? The answer is that the adults seem to have little trouble guessing the child’s modal intent, knowing the child intimately, and knowing the ongoing communicative context. To illustrate how this transpires, consider the following series of short modal interactions. Each opens with a short, rather elliptic initial modal move by the child, followed immediately by the adult’s interpretation — most often accurate — of the child’s speech-act intention; then often followed by the child’s confirmation of the adult’s modal interpretation. Thus (Eve-I): (7) a. EVE: Oh look, my pencil. (request?) MOT: There’s one in the kitchen (stating relevant facts) on the table counter. There’s one in the kitchen. (stating relevant facts) You may have that one. (offer) EVE: Write another pencil. (confirmation of goal) (p. 1) b. EVE: Candy? (request?) MOT: Candy? I think not. (rejection) EVE: Candy. (reiterated request) MOT: You have animal crackers (counter offer) on the table. (p. 1a) c. EVE:

That Fraser pencil.

(statement of fact?)

. Bates et al. (1975/1979) deemed this issue problematic, suggesting that adults often misinterpret the child’s speech-act intention at an early stage. I find this to be, largely, not the case at the age range studied here (1;8–2;9).

COL: EVE:

The ontogeny of complex verb phrases 

Can you write? Yeah.

(epistemic-modal question) (epistemic confirmation) (p. 1a)

d. EVE: Mom napkin. MOT: Oh, d’you want a napkin too? There.

(request?) (offer)

e. EVE: Look Fraser napkin. COL: Yes. You’ve got one. EVE: There. (p. 2)

(statement of facts?) (agreement & added facts) (agreement on facts)

f. EVE: MOT:

(request?)

Fraser blow nose, blow nose. Wipe your nose? Can you blow? That’s a good girl.

(question on modal intent) (counter offer) (reward for compliance) (p. 3)

g. EVE: FAT: EVE:

Sit Pop lap. You don’t want to sit on my lap now. Tomorrow. Morrow. (p. 3)

(request?) (rejection) (counter offer) (acceptance of alternative)

h. EVE: I put sugar in it. (offer?) MOT: I had sugar in my coffee. (incompatible facts) I don’t need any more sugar. (decline offer) (p. 4) i. EVE: I brush-ing. COL: What are you doing? EVE: [???] brush-ing. (p. 5)

(statement of facts?) (question of facts) (re-statement of facts)

j. EVE: MOT: EVE: MOT:

(uninterpretable utterance) (request for interpreted) (request?) (offer)

[???]. Do what? Self. What? Oh, you want one yourself?

 T. GivÓn

EVE: Eve get a Kleenex. MOT: Alright, take one.

(restated request) (offer) (p. 8)

k. EVE: Fall down. MOT: I know you fell down. EVE: That mine. (p. 17)

(statement of facts) (epistemic amplification of facts) (topic change)

l. EVE: [???] fall. MOT: It fell? I don’t know whether it did. EVE: It [???] fall down. Fall down Be a horsie. MOT: Be a horsie. Okay. EVE: Be a clip-clop. (pp. 47–48)

(statement of facts?) (question of epistemic intent) (epistemic uncertainty) (restatement of fact?) (topic & modality shift; request?) (granting the request) (re-stating request)

m. EVE: MOT: EVE: MOT:

(statement of facts?) (question of facts) (restatement of facts) (agreement on facts)

Baby. What’s Eve doing? Carry-ing a baby. Yeah. (p. 43)

On the relatively rare occasion when the adult’s interpretation of the child’s modal intent is unacceptable to the child, negotiations may ensue, and may proceed till the issue is resolved. Thus consider (Eve-I): (8) EVE: Get a spoon. (request?) MOT: Hmm? (incomprehension) EVE: Got a spoon. (statement of fact?) MOT: I forgot a spoon? (epistemic interpretation) No, you don’t get a spoon. (deontic interpretation; refusal) You don’t need one. (refusal) (p. 4) More often in cases of confusion, the adult responds with a question to clarify the child’s modal intent. This may be seen in (7f,g,h,k,l) above, as well as in (Eve-II): (9) EVE: Fraser … Fraser [???] top. COL: What do you want me to do?

(request?) (question about intent)

The ontogeny of complex verb phrases 

EVE: Take the top [off]. (re-statement of request) Fraser open my tinker-toy [box]. (re-statement of request) COL: Okay. (granting request) (p. 24) As I hope to show later on, a fine-grained qualitative analysis of these modal interaction units reveals multiple instances where complex modal-grammatical expressions are assembled collaboratively, across child-adult or adult-child conversational turns.

4. What counts as complex modal construction? In his study, Diessel (2005) was rather strict about what counted as a complex verb-complement construction in the child. Consequently, several complex constructions that fit the V-complement syntactic pattern were not included; most conspicuously: (10) a. Serial-verb constructions:

Let’s go (and) have supper. Come (and) get it.

b. Cognate-object constructions:

Have a drink/a seat. Take a nap/a bath/a turn. Make a mistake/a bad judgement. Give a lecture/a massage. Get a haircut/satisfaction

For the sake of completeness, the uses of such constructions by both the child and adult were included in this study. A more pressing reason for expanding the range of relevant constructions involves the fact that almost all deontic and epistemic verbs that take clausal complements also take simple nominal objects (Dixon 1991; Givón 1993b, 2001). What is more, in both language diachrony (Heine & Kuteva 2007; Givón, in this volume) and language ontogeny, using these verbs with nominal objects precedes their use with verbal complements. At least in principle, then, such simplex uses of deontic and epistemic verbs in child language ought to be counted as potential precursors of the complex verbal construction. What I hope to document here is that many modal MIUs reveal a rather flexible boundary between the simplex and complex uses of such verbs in both child and adult. I will illustrate this with a few examples from the Nina transcripts. Our three children use ‘want’ and other modal verbs with nominal objects (‘want-NP’) long before they use them with verbal complements (‘want-to-VP’). However, in almost all the early examples of their use of ‘want-NP’, a clear verbal interpretation of the nominal construction is possible, and indeed natural. Thus, in

 T. GivÓn

the Nina-I transcripts, 20 instances of ‘want-NP’ were recorded, as against only 3 of ‘want-to-VP’. As illustrations, consider (Nina-I): (11) a. ‘want’-NP:

Want toast. (⊃ ‘to eat’; p. 28) Want juice. (⊃ ‘to drink’; p. 28) Toast coming. I want it. (⊃ ‘to eat’; p. 30)

b. ‘want’-to-VP:

Wanna get down. (p. 49) Want hug. (p. 51) I want it hug. (p. 51)

The last blend construction in (11b) is also found as an early stage of the diachronic development of verbal complements (Givón 1991d; Heine & Kuteva 2007; Heine 2008; see ch. 4). The text-frequency disparity between the nominal and clausal complement is somewhat reduced in the Nina-II transcripts, albeit with too small a sample: 4 ‘want’-NP vs. 2 ‘want-to-VP’. And it is further reduced in Nina-I transcripts: 36 ‘want-NP’ vs. 21 ‘want-to-VP’. The same can be said of epistemic verbs such as ‘know’, ‘see’ or ‘look’, although here the overall text-frequencies are much lower (Diessel 2005). Thus for example, in the Nina-I transcripts, one finds 10 instances of the child’s use of ‘look-at-NP’ or ‘look-here/ there’, but only 2 instances of ‘look’ associated with a clausal complement (Nina-I): (12) a. ‘look’-at-NP:

Look at Mommy. (p. 49) Look at dolly book. (p. 49) Look at dolly book. (p. 49) Look at dolly book. (p. 49) Look at dolly book here. (p. 49) Look at this. Dolly book. (p. 49) Look at dolly book here. (p. 49) Look at dolly. (p. 49) Look here. (p. 49) Look. This way. (p. 49)

b. ‘look’-S:

Look. Drink a dolly. (p. 42) Look here’s Mommy book. (p. 49)

In either complement form in the early-stage transcripts, ‘look’ serves as a grammaticalized epistemic direct speech-act marker of directing attention. The same distributional tendencies are observed with ‘see’, with the bulk of the examples involving the same epistemic direct speech-act function of directing attention. Thus in the Nina-II transcripts, we find 3 examples of the child’s use of ‘see’-NP and only 1 of ‘see’-S (Nina-II):

The ontogeny of complex verb phrases 

(13) a. ‘see’-NP:

Let Snoopy see him. (p. 17) Oh, you want to see it. (p. 25) You see that in there? (p. 32)

b. ‘see’-S:

See Snoopy has those feet.

(p. 17)

Finally, in the Nina-III transcripts, we find not a single instance of the child’s use of ‘see-NP’, but 4 of ‘see-S’ in its various versions, again with the same epistemic speech-act function of directing attention (Nina-III): (14) ‘see’-S:

See what this is. (p. 14) See they knock the tree down. (p. 49) Oh, see they move. (pp. 15–16) And a ribbon in her hair. See. (p. 43)

Similar considerations can be applied to verbs such as ‘have’, ‘make’, ‘take’, ‘get’, ‘go’, ‘come’ and others, which can be used as auxiliary main verbs in complex modal constructions, but still appear at higher frequencies with nominal or prepositional objects in the early stages. Their early use with nominal objects is again, at least potentially, a developmental precursor to their later complex-VP use, in language ontogeny as in diachrony. For these reasons, I have elected to err on the side of inclusion, counting — in both adult and child — all the instances of verbs that can become modal operators over verbal complements.

5. The communicative context: A quantitative analysis In this section I will present four quantitative measures that probe into the general adaptive — communicative, functional, contextual — characteristics of the child-adult modal interactions studied here. In the main, this opening foray into our conversational texts reveals the essential soundness of the way Diessel and Tomasello characterize the early child use of modal expressions. What it also reveals, however is that the adult interlocutor is using the very same modal structure as the child.

5.1 Who takes the initiative for launching modal interaction? As noted earlier, each of our MIUs is launched by either the child or the adult; and either can initiate modality change in mid-interaction. It was thus of interest to see who takes the initiative in launching a new modality. For this purpose, modalinitiative gestures were divided into the two broad general types, deontic-manipulative and epistemic-informative. Under deontic-manipulative, I counted all direct here-and-now manipulative speech-acts, as well as expressions of intension-to-act

 T. GivÓn

in the immediate future. The latter may be considered, at least in the early-stages of the CHILDES transcripts, a species of promise or, occasionally, warning or threat. For example, in (15a,b,c) below the modal use of ‘would-like’, ‘can’t’ and ‘want’ are clearly manipulative. But so are the uses of ‘be-gonna’ and ‘will’ in (15d,e,f). Thus (Eve-II): (15) a. MOT: Would you like a graham cracker? (offer) EVE: Yeah. (p. 1) (acceptance of offer) b. EVE: Sue, put my sweeper down. (request) MOT: Can’t you do it? (rejection & counter-request) EVE: No. (p. 9) (refusal) c. EVE: COL: EVE:

Cromer … Fraser sit in chair. (request) Do you want me to sit (offer) over there? In the chair. (p. 15) (reconfirmed request)

d. MOT: Are you gonna sit (request/invitation) at the table? EVE: No. (p. 1) (refusal) e. EVE: MOT: EVE: MOT:

That my box. Look that? (protest; directing attention) I’m goin’ to steal your box. (immediate intent; threat?) What do-ing, Mom? (question of facts; alarm?) I’m going to use (immediate intent) your box. (p. 5)

f. MOT: EVE: MOT: EVE:

You lost two of them. (statement of fact; blame) [???] lost two. (re-statement of facts) I think I’ll just cut (manipulation, warning) that off, Eve. It’ll be easier. (softened manipulation) Wait a second. (manipulation) Think [???] cut (echo of warning; promise?) that off. (p. 6–7)

Under epistemic-informative I grouped ‘present/progressive’, ‘past/perfect’ and ‘non-immediate future’. In the children’s speech, the distinction between ‘present’ and ‘progressive’, or ‘past’ and perfect’, is not easy to demonstrate, due to lack of grammatical marking. One could of course maintain that the context — the adult’s directly-preceding turn — disambiguates the distinction. Thus consider:

The ontogeny of complex verb phrases 

(16) a. MOT: EVE: MOT: EVE:

Who is that? (Q-PRESS) That Jim. (PRESS) What’s he do-ing? (Q-PROG) Jump-ing (Eve-II, pp. 47–48) (PROG)

b. MOT: What is Mommy do-ing? (Q-PROG) NIN: Fix a dolly. (PROG?) MOT: Is she fix-ing up dolly? (Nina-I, p. 36) (Q-PROG) In (16a), the adult ‘present’ question is answered with the child’s ‘present’ (unmarked) form, while the adult’s ‘progressive’ question is answered with the child’s marked ‘progressive’ form. In (16b), the adult’s ‘progressive’ question is answered by the child’s unmarked ‘present’ form, which at this stage in Nina’s speech may mark either present, past, future or progressive. In context, however there is no reason to assume that the intended meaning was not ‘progressive’. For the purpose of the current measurement, the difference between ‘present’ and ‘progressive’ is not all that important, given that both are sub-species of here-and-now, non-displaced temporal reference. Either initiating an MIU or initiating a mid-MIU modality change were counted as taking modal initiative. Thus in (17) below, the child initiates the interaction and the same modality is maintained by both interlocutors throughout: (17) NAO: MOT: NAO: MOT: NAO: MOT: NAO:

Fix. (request) You can do it, honey. (manipulation) You just have to be patient. (manipulation) Fix. Fix. (repeated request) Oh, get it in the right place (manipulation) and then you can do it. (manipulation) Fix. (repeated request) Get it in the right place. (manipulation) You don’t want to break it. (warning) Fix. (Naomi-I, p. 22) (request)

In (18), on the other hand, the child initiates the interaction in a deontic mode (18a), the mother shifts to the epistemic (‘perfect’) in (18c), then immediately back to the deontic in (18d). The child then shifts to the epistemic in (18e) and then back to the deontic in (19f), which is maintained to the end of the interaction (Naomi-I): (18) a. NAO: More juice. b. MOT: More juice? c. The juice is almost gone.

(request) (clarification of request) (EPIST, PERF)

 T. GivÓn

d. Want some vitamins, Naomi? e. NAO: All gone. All gone. f. More vitamin. g. MOT: Wait. h. NAO: Sit. Juice. Mommy. I want it. (Naomi-I, p. 27)

(offer) (EPIST, PERF) (request) (manipulation) (manipulation)

Tables 1–3 below summarize the overall results of who takes the initiative and in what modality, for each child at all three developmental stages.

Table 1. Modal initiator: Eve-I-II-II. Deontic

Epistemic

(IMM FUT) PROG/PRES PAST/PFV Initiator I: adult child II: adult child III: adult child

FUT

Total-E

Total

N

%

N

%

N

%

N

%

N

%

N

%

31 55

68.8 65.4

11 21

24.4 25.0

3 9

6.6 10.7

/ /

0.0 0.0

14 30

31.2 34.6

45 84

100.0 100.0

22 41

56.4 54.6

12 28

30.7 37.3

4 5

10.2 6.6

1 1

2.5 1.3

17 34

43.6 45.4

39 75

100.0 100.0

41 40

60.2 55.5

16 29

23.5 40.2

10 3

14.7 4.1

1 /

1.4 0.0

27 32

39.8 44.5

68 72

100.0 100.0

Table 2. Modal initiator: Naomi-I-II-II. Deontic

Epistemic

(IMM FUT) PROG/PRES PAST/PFV Initiator I: adult child II: adult child III: adult child

FUT

Total-E

Total

N

%

N

%

N

%

N

%

N

%

N

%

58 29

53.2 34.9

39 50

35.7 60.2

10 4

9.1 4.8

2 /

1.8 /

51 54

46.8 65.1

109 100.0 83 100.0

17 66

33.3 49.3

23 65

45.1 48.5

11 3

21.5 2.2

/ /

/ /

34 68

66.6 50.7

51 100.0 134 100.0

30 42

34.1 46.1

29 35

32.9 38.4

20 13

22.7 14.2

9 1

10.3 1.3

58 49

65.9 53.9

88 100.0 91 100.0

The ontogeny of complex verb phrases 

Table 3. Modal initiator: Nina-I-II-III. Deontic

Epistemic

(IMM FUT) PROG/PRES PAST/PFV Initiator I: adult child II: adult child III: adult child

N

%

N

%

46 41

46.4 39.4

46 58

46.4 55.7

48 88

35.8 64.2

74 46

88 49

41.8 45.8

102 43

N

FUT

Total-E

Total

%

N

%

N

%

N

%

6 5

6.1 4.8

1 /

1.1 /

53 63

53.5 60.5

99 104

100.0 100.0

55.2 33.5

10 3

7.4 2.1

2 /

1.4 /

86 49

64.1 35.7

134 137

100.0 100.0

48.3 40.1

19 11

9.0 10.2

2 4

0.9 3.7

123 58

58.2 54.2

211 107

100.0 100.0

While the data is not amenable to inferential statistics, it appears that neither subject (Eve, Naomi, Nina) nor developmental stage (I, II, III) nor diadic participant (child vs. adult) reveal a trend. Overall, the child and adult seem to stand roughly on a par, first in terms of who takes the initiative, and second in terms of the balance between deontic and epistemic modal goals of the interaction.

5.2 Spatio-Temporal displacement Early communicative modes, be they those of animals, 2nd language pidgin or early child language, are notoriously anchored in the intimate referential universe of hereand-now, you-and-I, this-and-that visible on the scene. (Carter 1974; Bates 1976; Bates et al. 1975/1979). Tables 4–6 below document this vividly about child communication at this early stage. Taking all MIUs in each transcript, all utterances (‘clauses’) bracketed by a period [.] were counted, including one-word utterances (except yes/no). The latter elliptic interjections take their semantic valuation anaphorically from the preceding utterance, and would not have significantly changed the overall results. All deonticmanipulative utterances were counted as ‘immediate-future’. And again, the difference between ‘progressive’ and ‘present’ for the child is not altogether reliable. In all three subject-diads, at all three stages, the temporality is predominantly here-and-now. This is consonant with Diessel and Tomasello’s observation that the child’s modal grammatical devices are used, overwhelmingly, to mark direct speechacts. But in two of the children there seems to be a drop in the percent of here-andnow temporality in the last stage (III). Most important, across diads and stages, the adult discourse is just as here-and-now oriented as the child’s. This is consonant with the suggestion that grammaticalized modal expressions are used by the adult, at least in these transcripts, in very much the same way as they are by the child. Though the adults may be adjusting to the children (down-shifting).

 T. GivÓn

Table 4. Temporal displacement: Eve-I-II-III. Non-displaced (here & now) PRES I: adult child II: adult child III: adult child

PROG IMM-FUT

TOT

Displaced PAST PFV FUT TOT

% here & now

53 21

53 51

197 111

303 183

22 19

6 1

9 0

37 21

89.1% 89.7%

94 75

29 25

94 101

217 201

22 14

16 7

7 8

45 29

82.8% 87.3%

100 38

26 20

166 136

292 194

38 21

11 4

9 5

58 30

83.4% 86.6%

Table 5. Temporal displacement: Naomi-I-II-III. Non-displaced (here & now) PRES PROG I: adult 89 child 115 II: adult 50 child 85 III: adult 74 child 47

Displaced

IMM-FUT

TOT

PAST PFV FUT TOT

% here & now

66 53

134 98

289 266

8 1

2 3

11 3

21 7

93.2% 97.4%

54 93

84 190

188 368

20 5

/ /

/ /

20 5

90.3% 98.6%

36 40

121 144

231 278

50 26

2 4

27 13

79 43

74.5% 86.6%

Table 6. Temporal displacement: Nina-I-II-III. Non-displaced (here & now) PRES PROG I: adult child II: adult child III: adult child

Displaced

IMM-FUT TOT PAST PFV FUT TOT

% here & now

131 160

124 29

161 113

416 302

17 7

2 4

/ /

19 11

95.6% 96.4%

163 114

52 22

178 224

393 360

22 8

3 3

10 3

35 14

91.8% 96.2%

193 111

63 48

157 146

413 305

50 35

4 /

15 17

69 52

85.6% 85.4%

The ontogeny of complex verb phrases 

5.3 Speech-act value We have already noted the high prevalence of deontic interactions in our transcripts. Next, we focus more narrowly on the speech-act value of all utterances, dividing them between those that carry a deontic-manipulative intent and those that have an informative — either declarative or interrogative — intent. This determination is not always bound by grammatical marking. As noted earlier above, the child’s utterances at these early stages are often elliptic and grammatically unmarked. So that their modal intent is determined — by the adult interlocutor as well by the researcher — from the immediate discourse context. Further, many of the adult’s manipulative gestures are so-called indirect speech-acts, using either the declarative or interrogative grammatical form. Tables 7–9 below summarize the numerical results.

Table 7. Speech-act distribution: Eve-I-II-III. Speech act Deontic

Epistemic

Manip. N I: II: III:

ADU: CHI: ADU: CHI: ADU: CHI:

175 124 35 63 78 104

Declar %

N

47.5 50.0 14.3 24.9 21.1 35.3

102 119 120 169 161 148

Question %

27.7 47.9 49.1 66.7 43.5 50.5

N 91 5 89 20 130 42

Total

%

N

%

24.4 2.1 36.6 8.4 35.5 14.5

368 248 244 253 369 294

100.0 100.0 100.0 100.0 100.0 100.0

Table 8. Speech-ACT distribution: Naomi-I-II-III. Speech act Deontic

Epistemic

Manip. I: II: III:

ADU: CHI: ADU: CHI: ADU: CHI:

Declar

Question

Total

N

%

N

%

N

%

N

%

128 97 68 116 68 106

38.6 33.6 32.8 31.1 22.0 36.1

127 166 97 176 129 137

38.9 57.6 46.8 44.9 41.8 46.7

71 25 42 80 109 50

22.5 8.8 20.4 24.0 36.2 17.2

326 288 207 372 308 293

100.0 100.0 100.0 100.0 100.0 100.0

 T. GivÓn

Table 9. Speech-ACT distribution: Nina-I-II-III. Speech act Deontic

Epistemic

Manip. N I: II: III:

ADU: CHI: ADU: CHI: ADU: CHI:

155 110 181 215 131 110

Declar %

35.5 31.7 42.4 58.2 26.2 30.6

Question

Total

N

%

N

%

N

%

107 231 113 137 145 184

24.5 66.7 26.5 37.1 29.0 51.2

174 5 132 17 223 65

40.0 1.6 30.1 4.7 44.8 18.2

436 346 426 369 499 359

100.0 100.0 100.0 100.0 100.0 100.0

With much cross-diad and cross-stage variation, two trends seem to emerge out of these distributional data. First, within bounds, the child and adult use manipulative speech-acts at a similar rate, in the 20–50% range. And second, the children lag behind the adults, rather conspicuously, in producing interrogative speech-acts, although their use rises slowly toward the last stage (III).

5.4 Subject of modal expressions Another known characteristic of early childhood speech is that it is mostly about the speech-act participants — speaker and hearer. This is indeed strongly implicit in Diessel and Tomasello’s observations about the child use of epistemic and deontic main verbs as direct speech-acts, i.e., transactions between you and I. To demonstrate this, the subjects of all deontic or epistemic ‘higher verbs’ within all MIUs were counted and divided into 1st/2nd person vs. 3rd person. As we shall see later on, there is a strong correlation between reference to 3rd person subjects and the extension of main-verb usage from marking a direct speech-act to marking a description of an epistemic or deontic interaction. What is striking about our results, once again, is the virtual identity of the child and adult frequency-distribution patterns. Tables 10–12 below summarize the numerical distributions.

The ontogeny of complex verb phrases 

Table 10. 1st-2nd vs. 3rd pers. subject: Eve-I-II-III. Deontic 1–2 N I: adult 142 child 20 II: adult 62 child 32 III: adult 122 child 46

Epistemic

3

TOT

1–2

3

TOT

%

N

%

N

%

N

%

N

%

N

%

87.6 83.3

20 4

12.3 16.6

162 24

100.0 100.0

52 14

71.2 87.5

21 2

28.8 12.5

73 16

100.0 100.0

88.5 100.0

8

11.5 /

70 32

100.0 100.0

42 7

66.6 87.5

21 1

33.4 12.5

63 8

100.0 100.0

87.5 73.0

17 17

12.5 17.0

139 63

100.0 100.0

58 5

93.5 50.0

4 5

6.5 50.0

62 10

100.0 100.0

/

Table 11. 1st-2nd vs. 3rd pers. subject: Naomi-I-II-III. Deontic 1–2 I: adult child II: adult child III: adult child

Epistemic

3

TOT

N

%

N

%

N

123 53

86.0 98.1

17 1

14.0 1.9

143 54

54 94

94.7 92.1

3 8

5.3 7.9

96 104

88.8 98.1

12 2

11.2 1.8

1–2 %

N

3

TOT

%

N

%

N

%

100.0 110 100.0 14

68.7 60.8

50 9

31.3 39.2

160 23

100.0 100.0

57 102

100.0 100.0

82 42

82.0 66.6

18 21

18.0 33.4

100 63

100.0 100.0

108 106

100.0 111 100.0 66

84.1 88.0

21 9

15.9 12.0

132 75

100.0 100.0

Table 12. 1st-2nd vs. 3rd pers. subject: Nina-I-II-III. Deontic 1–2 N I: adult child II: adult child III: adult child

Epistemic

3 %

TOT

1–2

3

N

%

N

%

N

%

/ 1

/ 0.9

177 119

100.0 100.0

81 19

30.6 9.2

177 97

91.7 15 90.6 10

8.3 9.4

193 107

100.0 133 100.0 42

122 69

79.2 32 73.4 25

20.8 26.6

154 94

100.0 158 100.0 40

177 100.0 118 99.1

N

TOT %

N

%

183 186

69.4 90.8

264 205

100.0 100.0

63.0 64.6

78 23

37.0 35.4

211 65

100.0 100.0

59.3 43.9

107 51

40.7 56.1

256 91

100.0 100.0

 T. GivÓn

While there is again a considerable amount of fluctuation in the numerical values, two trends seem to emerge. First, the percent of 1st/2nd person subjects of epistemic verbs is almost always lower than of deontic ones. And second, the modal behavior of the child and adult is once again broadly similar. If one were to summarize the distribution of the adaptive-communicative features of complex modal expressions in our CHILDES transcripts, or at least in the modal interactions that form the immediate communicative context for produced complex modal expressions, one would have to agree with Diessel and Tomasello’s observation about the child’s modal behavior. But the adults behave, in this early-childhood communicative context, very much like the children. As we shall see further below, this observation extends to other facets of the use of complex-VPs.

6. Modality-marking grammatical devices 6.1 General considerations We come now to the structural core of this study — the classification and quantification of the modality-marking grammatical devices used by the child and adult in their modal interactions. While the rough division into deontic and epistemic holds in the main, it needs some refinement. The general division of modality-marking verbs into three major syntactic classes seems valid (Givón 2001, ch. 12; Diessel 2005). That is, for English: •• Modality verbs (‘want to do it’) take an equi-subject non-finite complement. •• Manipulation verbs (‘make someone do it’) take a nominal object-manipulee and an equi-object non-finite complement. •• Perception-cognition-utterance verbs (‘know that someone did it’) take a finite complement. Broadly, most deontic modal operators distribute in the first two groups, and most of the epistemic ones in the third. But many exceptions and refinements to this general classification must be noted. To begin with, we need to distinguish between the potential deontic speech-act use of a modal expression, and the actual use of such an expression as a direct manipulative speech-act. Consider for example (19a,b) below, where both the child and adult use ‘want’ as a direct-manipulative speech act (Nina-III): (19) a. MOT: Here’s another fence. NIN: Want another fence. [p. 25]

(offer) (request)

b. NIN: Many other fence? MOT: Want to build some more fences? [p. 25]

(request) (offer)

The ontogeny of complex verb phrases 

In contrast, in both (20a,b) the child and the adult use ‘want’ as a description of 3rd person volition. It is no accident that such usage is embedded in epistemic MIU contexts. Thus (Nina-III): (20) a. NIN: MOT:

Where, where can’t [= ‘can’] (FUT-HYPOTH) this go in the hole? Oh there. If they get out of there [,] (FUT-HYPOTH) these things. Oh, they want to get out of there. (PRES-VOLIT) What happened? [pp. 45–46] (PAST)

b. MOT: NIN: MOT: NIN: MOT:

What’s on the wall of the station? (Q-PRES) A apple. (PRES) No, that’s a clock. (PRES) Who are [???]? (Q-PRES) People want to know (PRES-VOLIT) what time it is. [p. 52]

The correlation between person (1st/2nd vs. 3rd) and speech-act value (direct vs. descriptive, respectively) is strong but not absolute. Thus, for example, in (21) below Nina uses a 3rd person subject of ‘want’ in a direct speech-act of request, a natural over-generalization from the much-more-common deontic-manipulative use of the verb (Nina-III): (21) NIN: MOT:

I forgot [to put] some more sticks in (PAST) this, in this … Two sticks wanna go in this truck. (request) Well, we’ll have to take some (manipulation) things out. [p. 47]

The same two-way modal potential is found in the use of modal auxiliaries by both the adult and the child. Thus in (22a,b), both use ‘can’ to mark a direct manipulative speech-act (Nina-III): (22) a. NIN: MOT:

A dog cookie. After he eats that one, can I, can I give him give him another one? Do you think he’ d like to eat another one? [p. 1]

b. NIN: A banana. MOT: Oh, can you make him eat a banana? [p. 3]

(request) (request) (Q-FUT/HYPOTH) (request) (manipulation)

 T. GivÓn

In (23a,b), on the other hand, both the child and the adult use ‘can’ as a description of ability, not surprisingly involving a 3rd person subject in an epistemic MIU context. Thus (Nina-III): (23) a. MOT: NIN: MOT: NIN:

What’s that? A circle. Is that the right place for it? Where can the other one go? [pp. 30–31]

(Q-PRES) (PRES) (Q-PRES)

b. MOT: NIN: MOT:

What’s he doing? Swing one one … What about this man? Do you think he can hang by your magnet? [p. 22]

(Q-PRES) (PRES) (Q-PRES)

(Q-PRES-ABIL)

(Q-PRES/ABIL)

A whole sub-class of modality verbs, the aspectual markers such as the progressive ‘be’, the perfect ‘have’, or the perfective ‘finish’ and ‘be (all) gone’, are clearly used only in an epistemic sense. The use of manipulation verbs in the CHILDES transcripts is a bit more consistent. Only two of those are used with any frequency, and they split down the modal line: ‘let’ (and at a much lower frequency ‘want’) is used by both the child and adult only in direct manipulative speech-acts, with 1st or 2nd person subject, as in (24a,b) below. In contrast, ‘make’ is used by both child and adult primarily as a description of manipulation/causation, as in (24c,d), even with 1st and 2nd person subject. Thus (Nina-III): (24) a. NIN: Yeah, let me give that to Poy now. I want … MOT: What do you want to do? NIN: I wanna give that to Poy now. [p. 1–2]

(request) (solicitation) (request)

b. MOT: Let’s set up a big village here. NIN: Okay, let’s do so. [p. 11–12]

(manipulation) (consent/request)

c. MOT: What did you do? NIN: I make the little bounce like a ball. I did it, Mommy. [p. 28]

(Q-PAST) (PAST, CAUS) (PAST)

d. NIN: Where’s the gas? MOT: Gas is what makes my car run. NIN: Oh. [p. 28–29]

(Q-PAST) (PAST, CAUS)

The same potential for double-usage exists in several perception-cognitionutterance verbs, most conspicuously ‘know’, ‘think’, ‘guess’, ‘say, ‘look’ and ‘see’. But since,

The ontogeny of complex verb phrases 

as Diessel (2005) has noted, these verbs are acquired much later, most of the uses in the CHILDES transcripts at the developmental stages studied here, by both child and adult, involves the direct speech-act of epistemic quantification of the complement clause, with 1st-2nd person subject. Most commonly, ‘know’ and ‘think’ are used in contexts of epistemic uncertainty or epistemic conflict. Thus consider (Nina-III): (25) a. MOT: NIN: MOT: NIN: MOT: NIN:

What colors are rabbits usually? (Q-PRES) Do you know? (Q-PRES, EPIST) Yup. (PRES) What color? (Q-PRES) Red. Blue. (PRES) No, they’re white. (PRES) Or red and blue. (PRES) You know that together, Mommy. [pp. 32–33] (PRES)

b. MOT: Is she gonna put what on her hair? NIN: Her ribbon on her hair. MOT: I don’t know. [p. 41]

(Q-FUT) (FUT) (PRES, EPIST)

c. NIN: Any more sticks? MOT: I don’t think so. [p. 18]

(Q-PRES) (PRES, EPIST)

d. MOT: Her ears are near her earrings, right? (PRES) NIN: Yup. Let me see. (PRES); (request) MOT: Oh, I guess she really doesn’t have ears. [p. 41–42] (PRES, EPIST) Many of these are also typical examples of cross-turn sharing of complex constructions. The perception verbs ‘look’, ‘see’ and ‘feel’ are used, at high frequency, as markers for the direct speech-act of directing attention. Again, most typically such usages involve a 2nd person subject (imperative form). Thus (Nina-III): (26) a. NIN: He’s, he’s eating a banana. (PROG) MOT: He is? My goodness. (PROG) NIN: Look at poy. [p. 3] (direct-attention) > (PRES) b. MOT: NIN: MOT: NIN:

What soft material. (PRES) Feel how soft it feels. (direct-attention) > (PRES) And her hair. (PRES) That’s a … (PRES) And a ribbon in her hair. See? [p. 43] (PRES) < (direct-attention)

 T. GivÓn

c. NIN: MOT: NIN: MOT:

The wheels don’t move. (PRES) No, I guess not. (PRES, CONCESSION) Oh see they move. (direct-attention) > (PRES) Oh, they do? [pp. 15–16] (Q-PRES)

d. NIN: Who goes in this little house? (Q-PRES) MOT: All the animals go in there. (PRES) See, this man is called Noah. (direct-attention) > PRES NIN: Oh. What is he doing with the animals? [p. 53] (Q-PROG) e. MOT: Look. What is the clown doing? (direct-attention) > (Q-PROG) Look, look at the clown, Nina. (direct-attention) > (PRES) NIN: Oh. MOT: Look at him. (direct-attention) > (PRES) See what he’s doing. (direct-attention) > (PROG) Can you see? (direct-attention) NIN: Yup. MOT: Look at the funny clown. (direct-attention) > PRES You don’t see him. (PRES; complaint of inattention) Look what I made him do? (direct-attention) > (PERF) See, Nina? (direct-attention) Look. What’s he doing? [p. 18] (direct attention) > (Q-PROG) All such uses of perception verbs, in spite of being themselves used as directmanipulative speech-acts, are embedded in highly epistemic MIUs. The later expansion of their use into descriptive epistemic modulation is driven, presumably, by their epistemic communicative context.

6.2 Quantitative analysis 6.2.1 Stage-I In Appendix-I, at the end of this chapter, examples are given of all the types of modal expressions used by the children and adults in our stage-I CHILDES transcripts. They are divided according to the various complement-taking verb types. Whenever applicable and attested, examples are given of both the direct speech-act and the descriptive use of modal expressions. The total number in the transcript of every usage type is given in parentheses. To make the modal intent of the examples

The ontogeny of complex verb phrases 

more transparent, they are cited with enough of the MIU context within which they are embedded. The frequency distribution of direct speech-act vs. descriptive use of modal expressions by the children and adults in our stage-I transcripts are summarized in Table 13 below. Table 13. Distribution of direct speech-act vs. descriptive use of modal expressions by children adult in stage-I transcripts. Child EVE: NAO: NIN:

Adult

Direct S-A

Descrip.

Total

Direct S-A

Descrip.

Total

8 (72.7%) 53 (89.8%) 26 (83.8%)

3 6 5

11 59 31

146 (91.2%) 136 (76.4%) 191 (89.6%)

14 42 22

160 178 213

In the absence of inferential statistics, and with some allowance for variation and low counts (Eve-I, Nina-I), the results, in the main, hold Diessel’s (2005) prediction about the child’s early use of complex modals expressions as direct speech-acts. However, they also show that the adult interlocutors’ use of these modal expressions falls very much in the same range as that of the child.

6.2.2 Stage II In Appendix 2, at the end of this chapter, all the usage-types of complex modal expressions in our stage-II transcripts, by both children and adults, are listed in a more compressed form. The frequency distribution of direct speech-act vs. descriptive use of these modal expressions by the children and adults in our stage-II transcripts is summarized in Table 14 below. Table 14. Distribution of direct speech-act vs. descriptive use of modal expressions by children adult in stage-II transcripts. Child Direct S-A EVE: NAO: NIN:

Descrip.

42 (93.3%) 3 (6.7%) 127 (95.4%) 6 (4.6%) 128 (83.1%) 26 (16.9%)

Adult Total 45 133 154

Direct S-A 94 (84.6%) 86 (87.7%) 215 (73.4%)

Descrip. 17 (15.4%) 12 (12.3%) 71 (26.6%)

Total 111 98 286

Again, a certain amount of cross-subject fluctuation persists. But the overall trend remains as in stage-I, with the bulk of usage in both children and adults conforming to Diessel’s (2005) characterization of the early child pattern — direct speech-act use of complex modal expressions.

 T. GivÓn

6.2.3 Stage-III In Appendix 3, at the end of this chapter, all the usage-types of complex modal expressions in our stage-III transcripts, by both children and adults, are listed in the same compressed form as for stage-II, above. The frequency distribution of direct speechact vs. descriptive use of these modal expressions by the children and adults in our stage-III transcripts is summarized in Table 15 below. Table 15. Distribution of direct speech-act vs. descriptive use of modal expressions by children adult in stage-III transcripts. Child Direct S-A EVE: NAO: NIN:

Descrip.

84 (91.3%) 8 (8.7%) 107 (85.2%) 15 (14.8%) 95 (66.4%) 48 (43.6%)

Adult Total 92 122 143

Direct S-A

Descrip.

Total

195 (83.5%) 30 (16.5%) 103 (61.9%) 62 (48.1%) 209 (67.8%) 99 (32.2%)

225 165 308

While inferential statistics is not available, it appears that in two of the children, Naomi and Nina, a shift towards higher use of the modal expressions in the descriptive sense begin at this stage — in both child and adult. Table 16, below, summarizes the frequency distribution of modal uses by the children and adults in all three stages, expressed in terms of the percent of direct speech-act use. Table 16. Percent of direct speech-act use of modal expressions stage. I

II

III

Subject

Child

Adult

Child

Adult

Child

Adult

Eve Naomi Nina

72.7 89.8 83.8

91.2 76.4 89.6

93.3 95.4 83.1

84.6 87.7 73.4

91.3 85.2 66.4

83.5 61.9 67.8

7. C ross-turn distribution of syntactic complexity: Paratactic precursors of complex verb phrases We come now to the crux of this investigation — the distribution of grammaticallymarked complex clauses across adjacent adult-child or child-adult turns, and the possibility that such distributed complexity is the paratactic precursor of the child’s complex VPs. What I present first below is a tentative typology of the various ways in which cross-turn collaboration between child and adult on the construction of complex modal expressions is transacted in our transcripts. I will begin by illustrating all

The ontogeny of complex verb phrases 

the types with examples from the Nina-II transcripts. In the following section, I will present the quantitative distribution of the types for all three subject-diads at all three developmental stages.

7.1 Q ualitative analysis: Types of cross-turn distributed complex modal expressions The following examples of the types of cross-turn distribution of complex modal expressions are taken from the Nina-II transcripts. I have divided them into two main categories: (i) types of the child’s responses to adult-initiated marked modal structures; and (ii) the adult’s response to two types of child modal expressions: (a) grammatically unmarked and (b) grammatically marked. For each category, I’ll give at least one deontic and one epistemic example. The response types are ordered from the least elaborate to the most elaborat — and eventually grammatically-marked.

7.1.1 C hild responses to grammatically-marked adult modal expression (Nina-II) (27) a. Appropriate yes/no elliptic responses EX: MOT: Would you like to play with the village? NIN: Yeah. [p. 5] EX: MOT: Do you think he’ll eat another one? NIN: Yup. [p. 1]

(offer) (accept) (Q-FUT) (FUT)

b. Response with an object of the complement clause EX: MOT: You want to give Poy a cookie? NIN: That one. [p. 1] EX: MOT: What is he eating? NIN: A dog cookie. [p. 1]

(offer) (accept) (Q-PROG) (PROG)

c. Response with an unmarked complement clause EX: MOT: What are you gonna do? NIN: Pat him. [p. 4] EX: MOT: Do you know what these are? NIN: What this thing? [p. 11]

(solicit) (intent) (Q-PRES) (Q-PRES)

d. Response with a marked complex modal construction EX: MOT: Can you make him do that? (manip.) NIN: I can’t do that. [p. 18] (refuse) EX: What is Poy doing? (Q-PROG) NIN: He’s eating a cookie. [p. 1] (PROG) EX: MOT: That would hurt, wouldn’t it? (Q-FUT/HYPOTH)

 T. GivÓn

NIN:

Yeah, on the road it would. [p. 48]

(FUT/HYPOTH)

e. Child-initiated marked complex modal construction with no adult prompt EX: MOT: Many little houses. (PRES) NIN: Let. Let’s put, let’s build these. [p. 13] (request) EX: NIN: Oh, oh, there will be another picture. (FUT) Just a minute. (request) See what this is. [p. 14] (direct-attention) EX: MOT: What’s on his pajamas? Oh. (Q-PRES) NIN: He’s hanging on two feet. [p. 21] (PROG)

7.1.2 Adult response to child’s previous modal turn (Nina-II) (28) a. Expansion: Marked complex-modal response to child unmarked expression EX: NIN: The cookie for Poy. (request) MOT: Do you want to give Poy a cookie? [p. 1] (offer) EX: MOT: What’s he eating? (Q-PRES) NIN: A banana. (PROG) MOT: Oh, can you make him eat a banana? [p. 3] (manip.) EX: NIN: What’s he doing? (Q-PROG) MOT: I don’t know. [pp. 14–15] (EPIST-PROG) EX: NIN: These wheels don’t move, Momma. (PRES) MOT: Oh, I think they’ll move. [p. 15] (EPIST-FUT) EX: NIN: Where does it belong? (Q-PRES) MOT: Where does it go? (Q-PRES) NIN: Yeah. (Q-PRES) MOT: I think it goes right here. (EPIST-PRES) NIN: Where? (Q-PRES) MOT: I don’t know … Maybe … (EPIST-PRES) I don’t know where it goes. (EPIST-PRES)

The ontogeny of complex verb phrases 

b. Reinforcement: Marked complex-modal response to child’s marked complex-modal expression: EX: NIN: I want … (request) MOT: What do you want to do? [pp. 1–2] (solicit) EX: NIN: Would you make a whole building? (request) MOT: Would I make what? [p. 53] (solicit) EX: This is gonna be a nurse. (FUT) MOT: Is that gonna be a nurse? (Q-FUT) EX: NIN: Cami doesn’t understand. (PRES) MOT: What doesn’t Cami understand? (Q-PRES) NIN: Doesn’t understand [how not to play] the rough. (PRES) MOT: You have to play gently, you mean. (PRES) NIN: Yup. [p. 3] (PRES) EX: NIN: He’ s eating that. (PROG) MOT: What is he eating? [p. 1] (Q-PROG) Of the five types of adult-child adjacent turn types in (27), the first three (27a,b,c) can be considered joint constructions of the complex modal structures, where the adult opens by contributing the marked modal structure and the child then contributes various chunks of the complement proposition — the gist of the communication — without an explicit modal marking. Only in types (27d,e) does the child contribute the full complex construction, in (27d) following adult prompting, in (27e) without prompting. It is significant, I believe, that the latter two types are either absent from or appear at low frequency in the Stage-I transcripts. Of the two types of child-adult adjacent turns in (28), (28a) is of course the most interesting kind of joint construction of a complex structure. The child contributes an unmarked, elliptic expression, which is then interpreted by the adult via modal expansion. Again, this type is more prevalent in the early stages, while type (28b), modal reinforcement, appears later.

7.2 Q uantitative analysis: Distribution of the various response types across diads and stages 7.2.1 C hild responses to marked modal expressions in the preceding adult turns The frequency distribution of the five types of child responses to modal expressions in the preceding adult turn is given in Table 17 below.

 T. GivÓn

Table 17. Child responses to marked modal expressions in preceding adult turn response type (27a–e). (a) Child-st.

N

EVE I II III NAO I II III NIN I II III

25 20 19 6 6 36 35 41 75

(b)

(c)

(d)

(e)

Total

%

N

%

N

%

N

%

N

%

N

%

32.0 19.2 16.5 5.6 4.7 19.5 28.9 24.2 30.0

12 10 22 36 7 13 33 30 33

15.3 9.5 19.1 33.9 5.5 7.0 27.2 17.7 13.2

36 19 12 35 6 15 51 28 12

46.1 18.2 10.4 33.0 4.7 8.1 42.1 16.5 4.8

5 19 26 12 25 20 / 21 23

6.6 18.2 22.6 11.3 19.6 10.8 0.0 12.4 9.2

/ 36 36 17 83 100 2 49 107

0.0 34.9 31.4 16.2 60.5 54.6 1.8 29.2 42.8

78 104 115 106 127 184 121 169 250

100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

7.2.2 A dult response to unmarked modal expressions in the child’s preceding turn The frequency distribution of the two types of adult responses to unmarked modal expressions in the preceding child turn is given in Table 18 below. Table 18. Adult response to unmarked modal expressions in preceding child turn

response type (28a–b). (a)

(b)

Total

Child

Stage

N

%

N

%

EVE

I II III I II III I II III

75 76 30 82 36 45 117 51 59

79.7 75.2 65.2 85.4 59.0 41.6 100.0 57.9 65.5

19 25 16 14 25 63 / 37 31

21.3 24.8 34.8 14.6 41.0 58.4 / 42.1 34.5

NAO

NIN

N 94 101 46 96 61 108 117 88 90

% 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

Because of the way the original modal interactions were selected (and within them the adjacent-pair types), and because of the low number of subjects and data points, it is not feasible to do inferential statistics on these numerical distributions. Still, several trends seem both present and plausible. i. In the adult-child adjacent turns, type (27a) should have no correlation to the developmental stage, since it is a perfectly universal elliptic response strategy to both epistemic questions and deontic suggestions. Types (27b) and (27c)

The ontogeny of complex verb phrases 

are the best candidates for early-stage syntax. Both show a drop in late stages. Types (27d) and (27e) are the end-products of learning, so their rise in the later stages is not surprising. ii. In the child-adult adjacent turns, type (28a) is the most striking early-stage cross-turn collaboration, where the child opens with an unmarked modal expression, and the adult then expands this by supplying the marked modal structure. It is thus not surprising that this type is most prevalent at the early stage, and decreases later on. Type (28b), adult reinforcement of a marked modal-expression produced by the child, is thus naturally a later development that depends on the child’s acquisition of marked modal structures. My aim in presenting the numerical distribution of all these types of interaction was to show the high prevalence in our early-stage transcripts of adjacent turns of type (27b,c) and (28a), where the complex modal structure is distributed across adjacent turns. In such modal interactions, the adult contributes most of the modal grammatical marking, and the child either responds with (27b,c), or contributes initially (28a), various grammatically-unmarked chunks of the complement clause, including the most elliptic yes/no responses (27a).

8. Child-adult usage comparisons To quite an extent, the results of this study uphold Diessel and Tomasello’s general thesis that in early child language the use of complement-taking main verbs is heavily tilted toward direct speech-act use, and thus that the main clause is not the semantic focus of the communication; rather, the complement clause is. But the results suggest that, at least in our early-stage CHILDES transcripts (age range 1;8 to 2;9), the adult’s use of complex modal expressions does not deviate significantly from that of the child. This brings us to our earlier assessment of the type of adaptive-communicative context we deal with at this developmental stage. As seen above, this context is strongly tilted towards non-displaced reference — here-and-now, you-and-I, thisand-that visible. It is also heavily invested in manipulative speech-acts. And it is largely speaker-hearer centered. This communicative context is characteristic of both pre-human and early-childhood communication. So it may be plausibly asked whether it isn’t this ‘primitive’ communicative context that motivates the adults’ child-like modal behavior. Are these adults in the CHILDES transcripts slumming or down-shifting? To assess this possibility, I selected as a comparison one chapter from recorded (and then transcribed) face-to-face oral narrative, the life-story of a retired rancher and oil-field worker. His story is, predominantly, about

 T. GivÓn

displaced time, place and referents, about past times, unfamiliar places, and people not known to his listener.6 The question we are concerned with is the narrator’s use of deontic and epistemic main verbs — to what extent does he use them as grammaticalized direct speech-act markers, the way both children and adults have done in our study? His past-time narrative is tilted heavily towards the epistemic, since his listener (myself) was interested primarily in his life story. But epistemic and deontic complement-taking verbs are found in both the narrative and the directquoted conversation portions of the texts. This affords us a revealing comparison between the two communicative contexts — within the same speaker. As an example of the use of epistemic and deontic modal operators in both the narrative and direct-quoted conversation portions of the text, consider: (29) And I knew I was gonna get so far so that I ever drive over there to see the family. So about this time this [oil] boom started here, see. Boy, I mean it was, I think a hundred and seventy-five rigs in here through most of the Fifties, y’know, drillin’ all this country up … So I came over here and started on this roughneckin’ job. But when I got here there was an old preacher up there that had seven, a little seventeen-acre place, and he’d been wantin’ to sell it and he ah, you know, he’d known him, I guess knew at least, kinda wanted family, he knew my dad. So he told me he said: “Harris you need a place”, said “let me sell you that little ol’ seventeen-acre farm …” It had a little three-room shack on it, see … It was up at Cedar, this side of Cedar Hill … That’s where my dad lived is up in that Cedar Hill area. So ah … he said: “I’ll sell it to you for fifty dollars down …” I said: “Well, OK, I’ll buy it”. And at that time, after we got all that movin’ done, y’know, I wasn’t making any money with the state … [p. 78] To further illustrate the high concentration of direct manipulative speech-act use of deontic modal expressions inside direct-quoted conversation, consider: (30) … He said: “Well” he said, “I’m gonna draw you a picture right here on this piece of paper, what you’re gonna find”. He said: “We gotta, we gotta come

. The narrative was tape-recorded over several long sessions in Bloomfield, NM in 1981–1982; when the speaker as ca. 62 years old. The text was then transcribed but not edited, with punctuation marks reflecting, as much as possible, the oral intonation units. For the narrative portion, the first 10 pages (70–79) were counted. For the inside-the-quotes portion, the whole 32 page chapter (70–101) was counted. For the text, the endless conversations, the winter trapping, year-round fiddlin’ and above all for the friendship, I am eternally indebted to Harris A. Brown (1923–1992). R.I.P.

The ontogeny of complex verb phrases 

out of that hole ’cause” he said, “this bit is wore out”. And he said: “I can’t get these other two guys to go up and I’m gonna have to have a man up there” he said. “Would you go up and try it for me as a favor?” he said, “’cause I got to come out” … [p. 73] Table 19 below offers a quantitative summary of the uses of modal operator in the narrative portion of text (10 pp. 70–79). Table 19. Distribution of modal uses in the narrative text Function Modality verbs ‘be-gonna’-VP ‘go’-to-VP ‘have-to’-VP ‘will’-VP ‘would’-VP ‘want’-to-VP ‘want’-NP ‘can’-VP ‘could’-VP ‘most-’ve’-VP ‘need’-NP ‘might’-VP ‘ought’-to-VP ‘try’-to-VP ‘come’-V (serial) ‘be-supposed-to’-VP ‘be-liable’-to-VP ‘threaten’-to-VP ‘be-liable’-to-VP ‘(’ve)gotta’-VP ‘start’-VP ‘gotta’-VP total-MOD: manipulation verbs: ‘tell’-NP-how-to-VP ‘tell’-NP-to-VP ‘tell’-NP-WH-VP ‘have’-NP-VP ‘let’-NP-VP ‘keep’-NP-VP

Direct (manipulative)

Descriptive

0 (0%)

10 2 11 14 4 11 1 7 17 1 1 1 1 5 1 1 1 1 2 4 1 2 99 1 1 2 5 1 1 (Continued )

 T. GivÓn

Table 19. (Continued) Function Modality verbs

Direct (manipulative)

Descriptive

‘get’-NP-(to)-VP ‘want’-NP-VP total-MANIP

0 (0%)

2 1 14

Epistemic verbs:

Epistemic quant.

‘think’-S S, ‘think’ ‘find-out’-S ‘know’-S S, ‘know’ S. ‘know’ ‘know’, S ‘know’-if-S ‘know’-WH/S ‘know’ (ellipsis) ‘guess’-S S, ‘guess’ ‘figure’-S S, ‘remember’ ‘see’-that-S ‘see’-if-S S, ‘see’ ‘See’, S ‘see’-WH/S ‘see’-NP-VP ‘tell’-NP-about-NP ‘tell’-NP: “…” ‘tell’-WH/S ‘tell’-if-S ‘say’: … “/” … “-’say’ ‘say’(,) S ‘mean’, S ‘figure-out’-(ellipsis) total EPIST

4 1

Descriptive 2 1

2 50 1 7 1 7 2 3 1 1 1

6

1 1 30 16 1 2

2 6 1 139 (63.1%)

1 1 1 10 2 2 53

81

The ontogeny of complex verb phrases 

Table 20 below offers the comparable distribution in the quoted conversational portions of the text. Table 20. Distribution of modal uses in the quoted conversation function Function Modality verbs ‘hate’-to-VP ‘go’-V (serial) ‘will’-VP ‘would’-VP ‘can’-VP ‘could’-VP ‘want’-to-VP ‘would-like’-VP ‘why-don’t-you’-VP ‘feel-like’-VP ‘need’-NOM ‘supposed-to’-VP ‘have-to’-VP ‘gotta’-VP ‘you-better’-VP ‘try’-VP

Manipulative 2 1 26 3 8 1 5 2 1

2 1

5 4 1 2 67 (80.7%)

manipulation verbs: ‘let’-NP-VP ‘want’-NP-VP total-MANIP

6 2 8 (100%)

‘know’-WH/S ‘know’-if-S ‘know’-S S, ‘know’ S. ‘know’ ‘say’-S ‘hear’-S ‘see’-NP-VP (raising) ‘tell’-NP-WH/S total EPIST:

5

1 1 1

total-MOD:

Epistemic verbs

Descriptive

Epistemic quant.

16

0 Descriptive

1 1 1 1 2 1 1 1 1 9 (90.0%)

1

While the data again does not allow inferential statistics, its main thrust is fairly dramatic. Narrative about the past, be it oral or written, is predominantly an epistemic (‘descriptive’) enterprise. This is because the main transactional goal

 T. GivÓn

in face-to-face narrative is descriptive-informative. It is thus not an accident that the bulk of epistemic verbs of belief, perception and utterance used in such narrative — 63.1% in this text — are used as direct epistemic speech-acts, i.e., as grammaticalized epistemic quantifiers on the complement clause. These epistemic speechacts are directed at the face-to-face listener. Likewise, it is not an accident that the modality and manipulation verbs that appear in narrative portion of the text are used — 100% — in their descriptive sense. Manipulation as a direct speech-act is not relevant in this communicative transaction, whose goal-posts have been set firmly in advance in the epistemic domain, and whose displaced temporality makes direct manipulative speech-acts infelicitous. In contrast, the direct-quoted face-to-face conversations had, so it appears, primarily deontic goals — to get things done. And the deontic modal operators used in that context change their valuation dramatically: They are used at the level of 80%-to-100% as direct manipulative speech-acts. The modal intent of complement-taking main verbs, it seems, has nothing to do with the child vs. adult developmental stage per se. Rather, it has much more to do with the communicative context. Of course, it just so happens that the communicative context of early childhood is predominantly here-and-now, you-and-I, non-displaced reference; and heavily tilted toward deontics (‘getting things done’) rather than epistemics (‘what is the state of the world’).

9. Conclusion 9.1 Child development and the communicative context In the early stages of modal development, children exhibit a strong tendency towards using modal operators — deontic and epistemic ‘higher’ verbs — as direct speech-act markers. But their adult interlocutors exhibit the very same trend in their face-toface communication with the children. This usage patterns seems strongly associated with the intimate communicative context of early childhood: here-and-now, youand-I, this-and-that visible. However, the same modal usage pattern is apparently found in oral face-to-face adult narrative, and in the quoted conversations embedded within it. In the non-conversational portion of the adult oral text studied here, deontic modal operators are not used as manipulative speech-act markers because the transaction goals in that context are predominantly epistemic. Once inside the quotation marks, the communicative goals shift toward the deontic, and the very same modal operators shift their use back to the deontic-manipulative.

The ontogeny of complex verb phrases 

It is of course yet to be determined which discourse type — or rather, which communicative context — is the true prototype of human language use. My own private bias, on evolutionary, diachronic and developmental grounds, leans strongly toward the developmentally-prior context of face-to-face oral communication. The modal usage that Diessel and Tomasello ascribed to early child language is thus not specific to children, but is rather the benchmark of face-to-face communication. While literacy and academic discourse are important cultural phenomena, I think it is best to view them — on evolutionary, diachronic and developmental grounds — as secondary, derived outcrops.

9.2 Semantics vs. syntax Diessel and Tomasello’s description of the two stages of child modal-use development pertains, strictly speaking, to semantic interpretation. There is no independent syntactic evidence that the two usages — direct speech-act vs. descriptive — differ syntactically. The semantic developmental trend observed by Diessel and Tomasello, while important, thus in no way supports the thesis that children expand simplex syntactic structures into complex ones. At most, the process involves a semantic shift — change of modal intent. The directionality of this change, however, is mediated by the communicative context, and it is practiced by both children and adults. What is more, the same directionality is seen in diachronic change in adults.

9.3 Combination and condensation: From parataxis to syntaxis The data of early child language suggest, strongly if not conclusively, that the earlier precursor of the child’s complex VP constructions, of whatever modality, is to be found in the joint construction of complex clauses across adjacent childadult or adult-child conversational turns. The later, ‘standard’ use of complex VPs by the child most likely emerges out of this paratactic precursor. This conforms closely to what has been observed in the diachronic rise of both complex VPs and complex NPs (Givón 2008, ch. 4,5). In both, earlier paratactic structures, with the two clauses packed under separate intonation contours, condense into later syntactic structures, with the two clauses falling under a single joint intonation contour. The main difference between the early-stage diachrony and early-stage ontogeny of complex syntax, in this respect, is that in diachrony the paratactic precursor for the condensation is found across two adjacent intonation units of the same adult speaker; while in early ontogeny, at least of complex VPs, the paratactic precursor for the condensed complex syntactic structure is collaborative, spanning adjacent turns of two different speakers.

 T. GivÓn

9.4 E arly childhood modal structure and the acquisition of Theories of Mind In both developmental and evolutionary terms, the shift in modal usage from direct speech-act to description of the modal states of others is part and parcel of several other major shifts in the communicative ecology (Givón 2008, ch. 12): •• the shift from here-and-now, you-and-I, this-and-that reference to displaced reference. •• the shift from manipulative to informative speech-acts. •• the shift from deontic modality to epistemic modality. •• the shift from mental models of one’s own mind to mental models of other minds. The close association — and temporal coincidence — of these four developmental trends is well attested in child language acquisition. The Theory of Mind literature, beginning with Premack & Woodruff (1978), suggests the development of deontic before epistemic mental representation, and the representation of the self ’s mind before the mind of others. This suggestion has received recent empirical support in the primate work of Cheney & Seyfarth (2007). The diachronic development of modal expressions suggests the same directional trend (Givón 1979; Heine et al. 1991; Hopper & Traugott 1993). The early stages of child development bear witness to the close association of the four developmental trends. Children acquire growing proficiency in the descriptive use of modal expressions roughly during the period — 2 to 4 years — when they acquire growing proficiency at Theory of Mind (Wellman 1990; Povinelli & DeBlois 1992; Meltzoff 1999, 2000; inter alia). The development noted by Diessel and Tomasello, of the child’s ability to use modal expressions descriptively, is thus part and parcel of the child’s growing ability to attribute epistemic and deontic mental states to others.

The ontogeny of complex verb phrases 

APPENDIX 1: Example and numerical distribution of direct speech-act vs. descriptive uses of complex modal expressions by the children and adults at stage-I 1A: EVE-I: Distribution of child use of complex modal expressions (a) Equi-subject modality verbs: ‘can’-VP: Descriptive (epistemic): (1) EX: MOT: And [when] Sarh’s a big girl, so can she. (FUT/ABIL) EVE: So can she. [p. 23–24] (FUT/ABIL) ‘(be)-gonna’-VP: Direct (manipulative): (1) EX: EVE: ‘Sue gon read Lassie’ (request; Sue =You) MOT: ‘I’m not gonna read Lassie’. [p. 9] (refusal) Descriptive (epistemic-future): (2) EX: EVE: She goin burp. (FUT) MOT: What? (Q) EVE: She’s goin burp. (FUT) MOT: She gonna burp. (FUT) She has to have milk first. [p. 14] (OBLIG) ‘have-(to)-VP: Direct (manipulative): (1) EX: EVE: Drink gain. MOT: After Sarah has a turn. EVE: Eve have it. MOT: Yes, you can have it, but you have to wait EVE: Have to wait. [p. 31]

(request) (deferral) (request) (permit) (oblig.) (oblig)

‘wanna’-VP: Direct (manipulative): (4) EX: MOT: You write on Eve’s paper. (manip.) EVE: No. (refuse) MOT: Look here’s a lot of paper … . (offer) EVE: Wanna write Fraser paper … [p. 36] (demand) (b) Equi-object manipulative verbs: ‘let’-NP-VP: Direct (manipulative): (1) EX: EVE: Get a stool. MOT: Get the cup, please and I’ll pour it. Bring the cup, eve. EVE: Let me have it. [p. 41] ‘help’-NP-(VP): Direct ( manipulative): (1)

(request) (manip.) (offer) (manip.) (request)

 T. GivÓn

EX:

EVE: Sue help Eve. MOT: Help Eve do what? EVE: Radiator. MOT: Oh, you wanna sleep on the radiator? [p. 44]

(c) Cognition-perception-utterance verbs:

(request) (solicit/offer) (request) (offer) (none attested)

1B: Eve-I: Distribution of adult use of complex modal expressions (a) Equi-subject modality verbs: ‘will’-VP: Direct (manipulative): (41) EX: EVE: Fraser wipe Eve nose ’gain. MOT: Come here. Mommy’ ll wipe your nose. [p. 5]

(request) (offer)

‘can’-VP: Direct (manipulative): (9) EX: EVE: Eve have it. MOT: Yes, you can have it … [p. 31] Descriptive (ability): (4) EX: MOT: And [when] Sarh’s a big girl, so can she. [p. 23–24]

(FUT) (FUT/ABIL)

‘may’-VP: Direct (manipulative): (5) EX: EVE: Look, oh, my pencil. MOT: There’s one in the kitchen. You may have that one. [p. 1]

(request) (PRES) (offer)

(request) (permit)

‘want’-NP/WH: Direct (manipulative): (11) EX: EVE: Napkin. (request) MOT: Oh, do you want a napkin too? [p. 2] (offer) ‘want’-VP: Direct (manipulative): (9) EX: EVE: Cracker on table. MOT: Oh, you want to have a cracker on the table? [p. 25]

(request) (offer)

‘would-like’-NP: Direct (manipulative): (2) EX: MOT: Would you like some fruit? EVE: No. [p. 64]

(offer) (refuse)

‘would-like’-to-VP: Direct (manipulative): (7) EX: MOT: Would you like to have your lunch now? EVE: No. [p. 14–15]

(offer) (refuse)

The ontogeny of complex verb phrases 

‘know-how’-to-VP: Descriptive (ability): (1) EX: EVE: Baby Sarah. MOT: She doesn’t know how to drink out of a glass. [p. 17]

(PRES) (PRES)

‘like’-NP: Direct (manipulative): (2) EX: MOT: Would you like to have some lunch? EVE: No. MOT: Papa will fix you one if you like. [p. 24] Descriptive (liking): (2) EX: MOT: Is that good? EVE: Yeah. MOT: D’you like it? [p. 22] ‘need’-NP: Direct (manipulative): (3) EX: EVE: [For]got a spoon. MOT: I forgot a spoon? No, you don’t get a spoon. You don’t need one. [p. 4] ‘supposed’-to-VP: Descriptive(?): (1) EX: EVE: That Fraser spoon. MOT: Thank you. What am I supposed to do with it?’ [p. 57] ‘try’-NP/VP/elliptic: Direct (manipulative): (2) EX: MOT: Not very good. No. EVE: I try again. MOT: Try again. Well, what are you trying to do? [p. 21] Descriptive: (1) (see directly above) ‘(be)gonna’-VP: Direct (manipulative): (8) EX: EVE: Sue read Lassie. MOT: No, Mommy’s not gonna read Lassie. [p. 12] Descriptive: (4) EX: EVE: She goin burp. MOT: She gonna burp. [p. 14]

(offer) (refuse) (offer) (Q-PRES) (PRES) (PRES)

(PAST)/(request?) (Q-PAST) (refuse) (refuse)

(PRES)

(FUT?)

(PRES) (offer/intent) (manip) (Q-PROG)

(request) (refuse) (FUT/imminent) (FUT/imminent)

‘go-and-V’ (serial): Direct (manipulative): (3) EX: FAT: You go eat your lunch. [p. 61] (manip)

 T. GivÓn

‘why don’t you’-VP: Direct (manipulative): (2) EX: MOT: Why don’t we have lunch? EVE: Drinking. [p. 20]

(manip) (request)

(b) Cognate object constructions: ‘have’-NOM: Direct (manipulative) contexts: (12) EX: EVE: Eve have drink of milk. MOT: After Sarah has a turn. [p. 29]

(request) (refuse)

(c) Equi-object manipulation verbs: ‘get’-NP-to-VP: Direct (manipulative): (3) EX: MOT: Is your grape juice all gone? (Q-PERFV) EVE: Yeah. (PERFV) MOT: Okay, let’s wipe your face then. [p. 7] (manip) ‘get’-NP-to-VP: Descriptive(?): (1) EX: EVE: [???] Eve ring. MOT: You don’t have a ring. When you get to be a lady, then you can have a ring. [p. 46]

(request) (PRES)/(refuse) (FUT)/(promise?) (FUT)

‘help’-NP-VP: Direct ( manipulative): (1) EX: EVE: Sue help Eve. MOT: Help Eve do what? EVE: Radiator. MOT: Oh, you wanna sleep on the radiator? [p. 44] ‘leave’-NP-VP: Direct (manipulative): (4) EX: MOT: You want me to smack you? EVE: No. MOT: Then put it away. Don’t touch it again Leave it lay right there. Leave it alone. [p. 50]

(offer)

(warning) (rejection) (manip.) (manip) (manip) (manip)

‘want’-NP-to-VP: Direct (manipulative): (2) EX: EVE: [???]. (request) MOT: Do what? (solicit) EVE: [???]-ing Eve. (request) MOT: What do you want me to do? [p. 45–46] (solicit) (d) Perception verbs: ‘see’-NP: Direct (attention): (4)

(request) (solicit/offer) (request)

The ontogeny of complex verb phrases 

EX: MOT: Where is the penny? EVE: [???] fall down floor. MOT: There it is. I see it, by the table. On the floor. See it? [p. 31]

(Q-PRES) (PAST) (direct-attention) (direct-attention) (direct attention)

‘see’-if-S: Direct (attention): (2) EX: MOT: Wanna go see if the coffee is read? (manip.; direct-attention) EVE: Yep. [p. 46] (consent) ‘see’. S: Direct (attention): (2) EX: MOT: See. She’s heavy. See. She’s heavy. [p. 43] ‘look-at’-NP: Direct (attention): (4) EX: EVE: Look, rocking-chair. MOT: It’s moving again. Look at the rocking chair. It’s doing it again. There is goes again. What is the rocking chair doing? [p. 4–5] (e) Cognition verbs: ‘know’-if-S: Direct (epistemic quantifier): (1) EX: EVE: [???] fall. MOT: It fell? I don’t know whether it did. [p. 47]

(direct-attention) (direct-attention) (PROG) (direct-attention) (PROG) (direct-attention) (Q-PROG)

(PAST) (Q-PAST) (EPIST-PAST)

‘know’-S: Direct (epistemic quantifier): (2) EX: EVE: Fall down. MOT: I know you fell down. [p. 17]

(PAST) (EPIST-PAST)

S, ‘think’: Direct (epistemic quantifier): (1) EVE: Eating bread too. MOT: She’s eating bread too, I think. [p. 57]

(PROG) (EPIST-PROG)

‘think’-S: Direct (epistemic quantifier): (2) EX: MOT: There’s a dog barking outside. Yeah. COL: I’m not sure. Yeah, I think it is. I’m sure it is. [p. 56)

(QUANT-PROG)

‘be-sure’-(S): Direct (epistemic quantifier): (2)

(see directly above)

(PROG)

1C: NAOMI-I: Distribution of child use of complex modal expressions

(a) Equi-subject (modality) verbs

 T. GivÓn

‘will’-VP: Descriptive (future): (1) EX: NAO: Daddy. (PRES) MOT: Daddy’s in Florida. In Florida. (PRES) He’ll be home tonight. (FUT) NAO: Daddy will be home tonight. [p. 19] (FUT) ‘can’-VP: Direct (manipulative)(?): (1) EX: NAO: Closed door. More. What’s this? What’s this? What’s this? Can’t get [it] off. Close door. FAT: Close the door. [p. 59] ‘want’-NP: Direct ( manipulative): (20) EX: MOT: Want some vitamins, Naomi? NAO: All gone. All gone. More vitamin. MOT: Wait. NAO: Sit. Juice. Mommy. I want it. [p. 27] ‘wanna’-VP: Direct (manipulative): (3) EX: NAO: Sit. Get up. Hug. Want hug. Want it hug. FAT: Do you want me to hug Georgie or Naomi? [p. 51]

(request) (request) (Q-PRES) (request?) (request) (MANIP.) (offer) (PERFV) (request) (manip) (request) (requests) (request) (request) (solicit)

‘go’-to-V: Descriptive (volition): (1) EX: NAO: Mouse tired. MOT: Oh, does he want to go to sleep? NAO: Go to sleep. MOT: Oh, I don’t think so. [p. 2]

(PRES) (Q-PRES/VOLIT) (PRES/VOLIT) (PRES/EPIST)

‘like’-NP: Descriptive (valuative): (2) EX: MOT: Naomi eat it. NAO: I like it. I like it. MOT: It’s good. NAO: No. [p. 32]

(manip.) (PRES/EVAL) (PRES/EVAL) (PRES/EVAL)

‘need’-NP: Direct (manipulative): (7) EX: FAT: How are you doing Nomi? (Q-PRES)/(solicit) NAO: Sugar. Need sugar. Need sugar on. (request) FAT: You need sugar? (offer)

The ontogeny of complex verb phrases 

NAO: Need sugar on. [p. 42]

(request)

‘need’-VP: Direct (manipulative): (1) EX: NAO: Leave it. I need cook. I need it. I need it. [p. 45]

(request) (request) (request)

(b) Equi-OBJ manipulation verbs ‘get’-NP: Direct (manipulative): (7) EX: NAO: Get it. Get it. Get. MOT: What are you getting, honey? NAO: Getting oof-oof. Getting woof. MOT: Getting off? NAO: Oof-oof. MOT: What are you getting? Are you going to get a doggie? NAO: Get doggie. [p. 17]

(request) (Q-PROG)/(solicit) (PROG)/(request) (Q-PROG; misinterpret) (clarification of request) (Q-PROG) (Q-intent)/(solicit) (request/intent)

‘get’-NP-LOC: Descriptive(causative)(?): (1) EX: NAO: Closed door. More. What’s this? What’s this? What’s this? Can’t get [it] off. Close door. FAT: Close the door. [p. 59] ‘leave’-NP-(ADJ): Direct (manipulative): (11) EX: NAO: Plate. FAT: An empty plate. NAO: Empty plate. Leave it. Leave it. Leave it. Leave it. Leave it alone. Leave it. Leave it alone. Leave it alone. Be careful. [p. 45–46] (c) Perception verbs ‘see’-NP: Direct (attention): (2) EX: NAO: Sun. Sun coming. Sun coming. Sun coming. Sun coming. MOT: Yeah. It is getting bright.

(request) (request) (Q-PRES) (request?) (request) (manip.)

(PRES) (PRES)

(manip.) (manip.) (manip)

(PROG) (PROG) (PROG)

 T. GivÓn

Sun. See sun. Where sun? [p. 24]

(direct-attention)

S. ‘see’: Direct (attention): (1) EX: NAO: Eating aspirin. Mommy see. (direct-attention)– (PROG) More Juice. More Juice. More Juice. (request) Eating. Eating. [p. 34] (PROG) ‘listen’ (ellipsis): Descriptive (perception): (1) EX: FAT: What are you doing with the sea-shell? (Q-PROG) Are you holding it over your ear? (Q-PROG) NAO: Holding hear. (PROG) FAT: Are you listening? (Q-PROG) NAO: Listen. [p. 50] (PROG) (d) Cognition verbs: ‘think’ (ellipsis): Descriptive (cognition): (1) EX: MOT: We’re not doing it. (PROG) I’m just thinking [of putting N. to bed]. (PROG) NAO: Thinking. (PROG) MOT: Thinking, yeah. With my head. (PROG) You think up there. [p. 20] (manip.) (e) Evaluative verbs: ‘feel’-ADJ: Descriptive (internal state): (1) EX: NAO: I feel better. (PRES)–(self-eval) MOT: Good. That’s good. (PRES)–(eval) I’m glad you feel better. [p. 35] (PRES)– (self-eval)–(eval)

1D: NAOMI-I: Distribution of adult use of complex modal expressions (a) Equi-subject modality verbs ‘will’-VP: Direct (manipulative): (7) EX: MOT: Do you want to comb your hair, Naomi? NAO: Comb hair. MOT: Mommy will get something for you to comb your hair. [p. 18–19] Descriptive (future): (7) EX: MOT: Sailboats. [looking at picture] NAO: Sailboats. MOT: We’ll see a sailboat this summer. [p. 9]

(offer) (request) (offer)

(PRES) (PRES) (FUT)

The ontogeny of complex verb phrases 

‘would’-VP: Direct (manipulative): (1) EX: NAO: That’s moon. (PRES) MOT: That’s not the moon, honey. (PRES) Would you please don’t push (request) your hands on the tray, honey? [p. 25] ‘can’-VP: Direct (manipulative): (16) EX: NAO: Toy doggie. MOT: Show me were it is. Can you point? [p. 7] Descriptive (ability): (4) EX: NAO: Home. MOT: The piggie didn’t want to stay home. See, sometime other people cry too because they have to stay home. Just like Nomi when she can’t go outside. [p. 12–13] ‘could’-VP: Direct (manipulative): (2) EX: NAO: Shadow. MOT: Shadow pictures. We could do shadow. [p. 15] ‘should’-VP: Direct (manipulative): (2) EX: NAO: Brush hair. MOT: You should wash your hair today. [p. 19–20] ‘might’-VP: Direct (epistemic): (1) EX: NAO: What’s this? MOT: I don’t know. I think it might be a matzo crumb too. [p. 32]

PRES/(request?) (manip) (manip) (PAST) (PAST) (HAB) (HAB) (HAB/ABIL) (request) (offer) (request) (manip.) (Q-PRES) (PRES)–(EPIST) (PRES)–(EPIST)

‘must’-VP: Direct (epistemic): (1) EX: MOT: Naomi, did you see (PAST) how the trees are blowing in the wind? (PROG) Must be windy. (PRES)–(EPIST) NAO: Windy. [p. 23] (PRES) ‘want’-NP: Direct (manipulative): (13) EX: MOT: Want some vitamins, Naomi? NAO: All gone. All gone. More vitamin. [p. 27]

(offer) (PERFV) (request)

‘want’-to-VP: Direct (manipulative): (6) EX: NAO: Fix.

(request)

 T. GivÓn

MOT: Get it in the right place. You don’t want to break it. NAO: Fix. [p. 22]

(manip.) (manip.) (request)

Descriptive (volition): (3) EX: NAO: Mouse tired. (PRES) MOT: Oh, does it want to go to sleep? [p. 2] (Q-PRES/VOLIT) ‘would-like’-to-VP: Direct (manipulative): (2) EX: FAT: Nomi, would you like to have some Famiglia this morning? (offer) NAO: Mmm mmm Mommy. [p. 41] (accept) ‘like-NP: Descriptive (liking): (3) EX: MOT: Did you like the matzo Nomi. NAO: I drop it. [p. 28] ‘like’-to-VP: Descriptive (future): (1) EX: MOT: The after that we could go over to school and go outside for a while. NAO: Yeah. MOT: Yeah, outside is where you like to be, isn’t it? [p. 20–21] ‘know-how’-to-VP: Descriptive (ability): (1) EX: NAO: Toy doggie. MOT: Show me were it is. Can you point? Do you know how to point? [p. 7] ‘need’-NP: Direct (manipulative): (1) EX: NAO: Juice. MOT: [to F.] I think she needs some aspirin. NAO: [???]. What ’s this? MOT: Aspirin. [p. 33]

(PAST) (PAST) (promise) (promise) (consent) (PRES)

PRES/(request?) (manip) (manip) (HAB/ABIL) (request) (manip.) (Q-PRES) (PRES)

‘try’-to-VP: Descriptive (attempt): (1) EX: MOT: Would you please don’t push your hands back on your tray, honey. (manip.) I’m trying to clean you off. [p. 25] (PROG) ‘(be)-gonna’-VP: Direct (manipulative): (1) EX: NAO: Woof-woof. MOT: What are you getting? Are you going to get a dogie? NAO: Get doggie. [p. 16–17] Descriptive (future ?): (2) EX: MOT: Watch. It’s going to pop.

(request) (solicit) (solicit) (request) (direct-attention) (FUT)

The ontogeny of complex verb phrases 

NAO: [???] hot. Toast coming. [p. 29] (FUT)

‘have-to’-VP: Direct (manipulative): (5) EX: NAO: Want it. Want it. MOT: It’s coming, Naomi. You have to wait till it pops out. [p. 29] Descriptive (obligation): (1) EX: NAO: Piggy crying. MOT: See the tears? Look at the tears. tion) That’s because the piggy had to stay home. [p. 12–13] ‘finish-up’-NP: Direct (manipulative context): (1) EX: NAO: More Famiglia. FAT: You’ve got a little bit more in there. You finish that up first. NAO: No more. [p. 44]

(request) (promise) (manip.) (PROG) (attract-atten-

(PAST/OBLIG) (request) (PRES) (manip.) (reject)

‘go’-to-V: Descriptive (motion): (1) EX: NAO: Mouse tired. (PRES) MOT: Oh, does it want to go to sleep? [p. 2] (PRES/VOLIT) ‘go’-(and)-V: Serial-verb (manipulative context): (1) NAO: Point. MOT: Point with your finger. See, like this. Go point. [p. 7]

(agree) (manip.) (direct-attention) (manip.

(b) Equi-object manipulation verbs ‘let’-NP-VP: Direct (manipulative): (8) EX: NAO: Oof-oof. [bringing a dog puppet] (PRES) MOT: Let’s make a shadow of that puppet, honey. [p. 17] (manip.) ‘have’-NP-VP: Descriptive (causative): (1) EX: MOT: How should we plan our day? Maybe we’ll have Naomi take a nap this morning. [p. 20]

(FUT?) (FUT/CAUS)

‘leave’-NP-ADJ: Direct (manipulative context): (1) EX: NAO: Leave it. Leave it. (manip.) FAT: Yeah, leave it alone, Nomi. [p. 45] (manip.) ‘want’-NP-VP: Direct (manipulative): (1) EX: NAO: I want it hug.

(request)

 T. GivÓn

FAT: Do you want me to hug Georgie or Nomi? [p. 51] (offer)

‘make’-NP-VP: Direct (manipulative): (3) EX: NAO: Comb hair. (request) MOT: Here’s a brush, Naomi. (offer) Make your hair feel good. [p. 18–19] (offer) Descriptive (causative): (1) EX: NAO: Where sun? (PRES) MOT: The sun is making it warm … [p. 24] (PROG/CAUS) ‘get’-NP-VP: Direct (manipulative context): (1) EX: MOT: Lie down on the floor so Mommy can get you dressed. NAO: No. [p. 39]

(manip.) (manip.) (refuse)

(c) Perception verbs ‘see’-NP: Direct (attention): (6) EX: MOT: Look, see the shadow. NAO: Shadow. [p. 20]

(attract-attention) (PRES)

Descriptive (perception): (5) EX: NAO: Where daddy? (Q-PRES) MOT: Daddy is working tonight, hone. (PROG) Daddy will be home tonight. (FUT) You’ll see him tomorrow morning. [p. 24] (FUT/PERCEP) ‘see’-if-S: Direct (attention): (1) EX: MOT: Let’s see if you remember all of them. Who’s this? NAO: Mr. Gum. [p. 9] ‘see’-WH/S: Descriptive (perception): (2) EX: NAO: What’s this? FAT: I can’t see what you’re pointing at. [p. 58] S. ‘see’: Direct (attention): (1) EX: NAO: [???]. (???) MOT: There it is. See. [p. 16] ‘see’, S: Direct (attention): (4) EX: NAO: What’s this? Man. [doing] MOT: See, this man is making shoes. [p. 62] ‘watch’-NP: Direct (attention): (1) EX: MOT: Hold it up nice.

(manip) (Q-PRES) (PRES) (Q-PRES) (PRES/PERCEP)

(direct-attention)–(PRES) (Q-PROG) (direct-attention)–(PROG) (manip)

The ontogeny of complex verb phrases 

Sit down and watch the shadow. [p. 18]

(direct-attention)

‘watch’. S: Direct (attention): (1) EX: MOT: Watch. It’s going to pop. (direct-attention)– (IMM. FUT) NIN: [???] hot. Toast coming. [p. 29] (PRES/PROG) ‘look-at’-NP: Direct (attention): (3) EX: NAO: Piggy crying. (PROG) MOT: See the tears? (attract-attention)> (PROG) Look at the tears. [p. 12–13] (attract-attention) ‘look’, S: Direct (attention): (1) EX: MOT: Look, see the shadow. NAO: Shadow. [p. 20]

(direct-attention)>(PRES) (PRES)

‘show’-DAT-NP: Direct (attention): (5) EX: MOT: Show me the mommy. (direct-attention) NAO: Mommy. [pointing]. [p. 5] (PRES) ‘show’-DAT-WH/S: Direct (attention): (1) EX: NAO: Toy doggie. MOT: Show me where is. [p. 7]

(PRES) (direct-attention)

‘hear’-NP-VP: Direct (attention): (2) EX: FAT: Do you hear the birds (direct attention)> singing? (PROG) NAO: [???]. (PRES) FAT: Yes, the sun is out. [p. 42] (PRES) ‘listen’ (ellipsis): Direct (attention): (1) EX: FAT: Are you listening? NAO: Listen. FAT: Yeah, listen. [p. 50]

(Q-PROG) (PROG) (direct-attention)

Descriptive (perception): (1) EX: FAT: Are you holding it over your ear? NAO: Holding ear. FAT: Are you listening? NAO: Listen. [p. 50]

(Q-PROG) (PROG) (Q-PROG) (PROG)

(d) Cognition verbs WH/S. ‘know’: Direct (epistemic quantifier): (1) EX: NAO: What’s this? MOT: I don’t know. [p. 32]

(Q-PRES) (PRES/EPIST)

 T. GivÓn

‘know’-WH/S: Direct (epistemic quantifier): (3) EX: NAO: Elbow. (PRES) MOT: Do you know where the elbow is? (Q-PRES/EPIST) NAO: Elbow. [pointing to picture] [p. 5] (PRES) ‘think’-S: Direct (epistemic quantifier): (7) EX: NAO: What’s this. MOT: I don’t know. I think it might be a matzo crumb. [p. 32]

(Q-PRES) (PRES/EPIST) (PRES/EPIS)

S, ‘think’: Direct (epistemic quantifier): (2) EX: NAO: Go to sleep. (request) FAT: I don’t think so. [p. 2] (PRES/EPIST)/ (refusal?) EX: NAO: What’s this? (Q-PRES) FAT: It’s a piece of foam I think. [p. 47] (PRES/EPIST) ‘think’ (about): Descriptive (cognition): (4) EX: MOT: Don’t cry. I’m thinking about it, honey. We’re not doing it. I’m just thinking. EVE: Thinking. MOT: Thinking, yeah. With my head. You think up there. [p. 20] ‘remember’-NP: Direct (attention): (2) EX: NAO: What’s this? Man. [doing?] FAT: See, this man is making shoes. Do you remember the other shoemaker? [p. 62]

(PROG) (PROG) (PROG) (PROG) (HAB)

(Q-PROG) (dir.-attention)/ (PROG) (dir.-attention)

‘remember’-WH/A: Direct (attention): (2) EX: MOT: Just like Nomi when she can’t go outside she cries (HAB) Remember how (direct-attention)/ you cry when you can’t go outside? [p. 12–13] (HAB) S. ‘understand’ (ellipsis): Descriptive (cognition): (1) EX: NAO: Hi.

(greeting)

The ontogeny of complex verb phrases 

MOT: Hi what? I don’t understand. [p. 8]

(PRES)

‘be-hard’-to-‘understand’: Descriptive (cognition): (1) EX: MOT: We’re making toast out of bread. (PROG) NAO: [???]. MOT: It’s kinda hard to understand. [p. 29] (PRES) ‘figure out’-WH/S: Direct (directive context): (1) EX: NAO: Daddy. Hi (greeting) FAT: Let’s figure out (dir. attention) what Nomi is going to wear today. [p. 53] (FUT) (e) Utterance verbs ‘say’-dir.quote: Directive contexts: (15) EX: NAO: Piggy sleeping. MOT: Piggy is sleeping. Can you say: “Piggy is sleeping”? [p. 12]

(PROG) (PROG) (mark dir. quote)

Descriptive (utterance): (2) EX: MOT: You don’t want that delicious honey. (manip.) NAO: Yes. (assent) MOT: You said: “Yes”. (PAST/DIR. QUOTE) You don’t mean a word of it. [p. 32] (PRES/EPIST) ‘tell’-DAT-WH/S: Directive context: (4) EX: NAO: Hat. (PRES) MOT: Tell me who this is. Peter. (EPIST. QUANTIFIER) NAO: Peter. [p. 9] (PRES)

1E: NINA-I: Distribution of child use of complex modal expressions (a) Equi-subject modality verbs: ‘go’: Descriptive (motion): (1) EX: MOT: Is the rabbit going fast? Uh? NIN: Go. MOT: It’s going. [p. 3] ‘like’-NP: Descriptive (liking): (2)

(Q-PROG) (PROG) (PROG)

 T. GivÓn

EX: NIN: Kitty cat. Big kitty cat. MOT: Do you like kitty cat? NIN: Like kitty cat. Like kitty cat. [p. 24] (b) Equi-object manipulation verbs ‘have’-NP-LOC: Direct (manipulative context): (2) EX: NIN: Duck room, have it o[n] wee. On the black. On the black. MOT: Are you going to put the duck in the black space? This is a puzzle. [p. 27] ‘make’-NP: Direct (manipulative context): (2) EX: MOT: Did you make the blocks fall down? NIN: Here. MOT: Uh? NIN: Make it, Mommy. MOT: You want me to make it? [p. 44]

(PRES) (Q-PRES) (PRES)

(request) (request) (solicit)

(Q-PAST) (request) (request) (solicit)

‘get’-NP: Direct (manipulative context): (9) EX: NIN: Get the ball. MOT: Get the ball? You want me to get the ball? NIN: Get the ball. [p. 54]

(request) (offer) (offer) (request)

‘take’-NP-LOC: Descriptive (removal): (2) EX: MOT: What am I doing? What is Mommy doing? NIN: Take it off. [???] off. MOT: Taking the pants off. NIN: Take off clothes. [p. 39]

(Q-PROG) (PROG) (PROG) (PROG)

(c) Perception verbs: ‘look’-(at)–(NP): Direct (attention): (11) EX: NIN: Open that. [book] (request) MOT: That doesn’t open. (PRES) That’s the end of the book. (PRES) Want to look at it some more? (offer) NIN: Look rabbit. [p. 18] (direct-attention/ request) ‘look’ [.]-S: Direct (attention): (2) EX: MOT: What are you giving dolly to drink? (Q-PROG) NIN: Look. Drink a dolly. [p. 42] (direct-attention)– (PROG) ‘fee’-ADV: Descriptive (internal): (2)

The ontogeny of complex verb phrases 

EX: MOT: Oh, you’re hugging the lady. Does she feel better? NIN: Feel better. [p. 60]

(PROG) (Q-PRES) (PRES)

1F: NINA-I: Distribution of adult use of complex modal expressions (a) Equi-subject modality verbs: ‘will’-VP: Direct (manipulative): (5) EX: NIN: Read. MOT: Won’t you read the bunny? NIN: Read the bunny. [p. 2] EX: NIN: The book. MOT: No, you can’t open that. It’ll tear. [p. 22]

(request) (manip.) (request) (request) (prohibit) (warn)

‘can’-VP: Direct (manipulative): (32) EX: NIN: Look, Mommy. (direct-attention/ request) MOT: Do you want me to take off your shoes too? (offer) Can you take off your shoe? (manip.) NIN: Hard. [p. 40] (complain) ‘shall’-VP: Direct (manipulative): (16) EX: NIN: Other kitty cat. MOT: Shall we find some other kitty cat? [p. 25]

(request) (offer)

‘have-to’-VP: Direct (manipulative): (3) EX: NIN: The book. (request) MOT: No, you can’t open that. It’ll tear. (prohibit) You have to just turn the pages. [p. 22] (manip.) ‘want-to’-VP: Direct (manipulative): (16) EX: NIN: More rabbit books. (request) MOT: Do you want to find another book with a rabbit in it? (offer) NIN: Here. [p. 19] (request) ‘would-like-NP: Direct (manipulative): (2) EX: NIN: Yummy. [eating a cookie] EX: MOT: Would you like some more cookies? [p. 38]

(PRES) (offer)

 T. GivÓn

‘would-like’-to-VP: Direct (manipulative): (2) EX: MOT: Want me to drink dolly’s milk? (offer) NIN: Yeah. (request) MOT: Oh, it’s so good. Umm. (PRES) Would you like to play with dolly’s milk? [p. 49] (offer) ‘like’-to-NP: Descriptive (liking): (7) EX: NIN: Panda. MOT: Do you like the panda? NIN: Yeah. [p. 8]

(PRES) (Q-PRES) (PRES)

‘like’-to-VP: Descriptive (volition): (7) EX: MOT: The guitar makes music, doesn’t it? (PRES) NIN: Yeah. (PRES) MOT: Do you like to sing? [p. 13–14] (Q-PRES) ‘try’-NP: Direct (manipulative): (3) EX: NIN: Hard. (PRES/complain) MOT: Is it hard to put it on? NIN: Here. MOT: You try it. Try again. [p. 27–28] ‘try’-to-VP: Direct (manipulative): (5) EX: NIN: Hard. MOT: It is hard? NIN: Yes. MOT: You try to take off your shoe. [p. 40] ‘(be)-gonna’-VP: Direct (manipulative): (11) EX: NIN: Read. MO: Won’t you read to bunny? NIN: Read to bunny. Read to bunny. MOT: Are you going to read to bunny? [p. 16] ‘go’-(LOC): Direct (manipulative context): (5) EX: MOT: Would you like to go out to supper with Mommy? NIN: Supper. [p. 56] Descriptive (motion): (2) EX: MOT: Is the rabbit going fast? NIN: Go. MOT: It’s going. [p. 3]

(Q-PRES) (request)

(PRES/complain) (Q-PRES) (PRES) (manip.) (request) (manip.) (request) (manip.)

(offer) (agree) (Q-PROG) (PROG) (PROG)

The ontogeny of complex verb phrases 

‘go’-(and)-V (serial): Direct (manipulative context): (3) EX: MOT: Go find the ball. Go find the ball. (manip.) Can you find the ball? [p. 52] (manip.) ‘why-don’t-you’-VP: Direct (manipulative): (1) EX: MOT: Oh dear, we have to start over again. (manip.) Oh, let’s start over again. (manip.) Why don’t you bring me the yellow block? (manip.) Where is the yellow block? (Q-PRES/manip.) NIN: Here. [p. 44] (PRES/comply) (b) Equi-object manipulation verbs ‘let’-NP-VP: Direct (manipulative): (22) EX: MOT: Look at the puzzle. Let’s take att the pieces out. NIN: Yeah. [p. 26–27] ‘make’-NP: Direct (manipulative context): (3) EX: NIN: Make it, Mommy. MOT: You want me to make it? Okay, let’s make it. [p. 44] Descriptive (construction): (2) EX: NIN: Here. MOT: You are making a building? NIN: Building. [p. 45]

(direct-attention) (manip.) (agree) (request) (offer) (manip.) (PRES) (Q-PROG) (PROG)

‘make’-NP-VP: Direct (manipulative context): (4) EX: MOT: Shall we make dolly dance? Let’s see, dance, dance. Make dolly dance. You make dolly dance. [p. 38]

(manip.) (manip.) (manip.) (manip.)

‘want’-NP-VP: Direct (manipulative): (11) EX: NIN: Untie. MOT: Want me to tie it? NIN: Off. Shoe off. [p. 41]

(request) (offer) (request)

‘get’-NP: Direct (manipulative context): (16) EX: NIN: Get big ball. Big ball. MOT: Shall we get the big ball? [p. 51]

(request) (offer)

 T. GivÓn

‘take’-NP-LOC: Direct (manipulative context): (4) EX: NIN: Books MOT: Look at the puzzle. Let’s take all the pieces out. NIN: Yeah. [p. 26–27]

(PRES/request?) (direct-attention) (manip.) (agree)

(c) Perception verbs: ‘see’-NP: Direct (attention): (3) EX: MOT: What’s the rabbit doing? NIN: Hopping. MOT: Uh-huh. And he’s painting too. See the rabbit? [p. 17–18]

(Q-PROG) (PROG) (PROG) (attract attention)

‘see’ (ellipsis): Direct (attention; manipulative context): (3) EX: NIN: Other kitty cat. (request) MOT: Shall we find some other kitty cat? (manip.) Let’s see. [p. 25] (manip.) ‘see’-if-S: Direct (attention; manipulative context): (3) EX: MOT: It’s hard? NIN: Yes. MOT: You try to take off your shoe. See if you can take it off. [p. 40]

(Q-PRES) (PRES) (manip.) (manip)

‘see’-NP-VP: Direct (attention; manipulative context): (1) EX: MOT: Oh, did it fall down? NIN: Yeah. MOT: Oh, can you build it some more? Let’s see you build it. [p. 43]

(PAST) (PAST) (manip) (manip.)

‘look’(at-NP): Direct (attention): (1) EX: NIN: Books. MOT: Look at the puzzle. Let’s take all the pieces out. [p. 26–27]

(request) (direct-attention) (manip.)

‘look’ (,) S: Direct (attention): (1) EX: MOT: Shall we build something? NIN: Oh, something. MOT: Oh, look let’s put all the blocks on top of each other. [p. 43]

(manip.) (request) (direct-attention/ manip.)

‘look-like-NP: Descriptive (perception): (1) EX: NIN: Big mouse, big mouse. MOT: He looks like a mouse, but he is a seal. [p. 16]

(PRES) (PRES/EPIST) (PRES)

(d) Cognition verbs

The ontogeny of complex verb phrases 

S, ‘think’: Direct (epistemic quantifier): (1) EX: NIN: A bird. MOT: That’s a bug, I think. Yes, that’s a bug. [p. 10]

(PRES) (PRES/EPIST) (PRES)

‘think’-S: Direct (epistemic quantifier): (3) EX: MOT: Do you think dolly is getting hungry again? NIN: Yeah. [p. 47–48]

(Q-PROG/EPIST) (PROG)

‘think’ (ellipsis): Direct (epistemic quantifier): (1) EX: MOT: He’s falling down. NIN: Yeah. MOT: Do you think so? [p. 3]

(PROG) (PROG) (PROG/EPIST)

WH/S, ‘remember’: Direct (epistemic quantifier): (1) EX: MOT: He’s playing the guitar. (PROG) NIN: [???]. MOT: Who plays the guitar, Nina? (Q-PRES/EPIST) Remember? ‘understand’ (ellipsis): Descriptive (cognition): (1) EX: MOT: What can you sing? (PRES/ABIL) NIN: Up down. (PRES) MOT: I don’t understand. [p. 13–14] (PRES) (e) Utterance verbs ‘say’-dir.quote: Direct (manipulative context): (9) EX: MOT: Can you say: “Giraffe”?

(manip.)

 T. GivÓn

NIN: Giraffe. [p. 8] vs. descriptive (comply) APPENDIX 2: Distribution of direct speech-act use of complex modal expressions by children ‘say’-WH/S: Descriptive (utterance): (2) and adults in stage-II transcripts EX: NIN: Meow, meow. 2 A: Distribution of child MOT: uses ofMeow? modalIspatterns inthe Eve-II that what cat says? (Q-PRES) NIN: Meow. [p. 9] Modality verbs Direct(manip.) Descriptive ‘have’-to-VP ‘gotta’-VP ‘can’-VP (you/I) better-VP ‘gonna’-VP ‘need’-NP ‘want’-NP ‘try’-to-VP ‘forget’-to-VP manipulation verbs ‘let’-NP-VP Perception-epistemic

13 2 1 2 6 4 4 1 /

/ / / / 1 / / / 1

4

/

Direct (attention)

Descriptive

4 /

/ 1

Direct (epist. quant.)

Descriptive

1 42 (93.3%)

/ 3

‘look’ ‘feel’ Cognition-epistemic ‘think’-S total:

2B: Distribution of adult uses of modal patterns in Eve-II Modality verbs ‘will’-VP ‘can’-VP ‘would’-VP ‘might’-VP ‘gonna’-VP ‘have’-to-VP ‘want’-NP ‘want’-to-VP ‘would-like’-NP ‘need’-NP ‘(had)-better’-VP Manipulation verbs ‘let’-NP-VP ‘want’-NP-VP

Direct (manip.)

Descriptive

20 7 1 / 6 3 1 1 5 5 5

7 2 1 1 2 / 1 / / / /

Direct (manip.)

Descriptive

3 3

/ /

Perception-epistemic ‘look’-WH/S ‘look’-like-NP ‘watch’-NP-VP ‘show’-NP-VP ‘hear’-NP ‘listen’ (ellipsis) ‘feel’-ADJ Cognition-epistemic ‘know’-if-S ‘know’-S ‘know’ (ellipsis) ‘remember’-WH/S ‘think’-S S, ‘think’ Utterance-epistemic ‘ask’-DAT (ellipsis) ‘say’-dir.quote total:

The ontogeny of complex verb phrases 

Direct (attention) 1 / 1 6 3 1 /

Descriptive / 1 / / / / 1

Direct (epist. quant.)

Descriptive

1 3 3 1 11 1

/ / / 1 / /

Direct

Descriptive

1 1 94 (84.6%)

/ / 17

2C: Distribution of child uses of modal patterns, Naomi-II Modality verbs ‘will’-VP ‘can’-VP ‘gonna’-VP ‘go-and-V’ (serial) ‘go’-LOC ‘want’-NP ‘wanna’-VP ‘stop’-(VP) ‘like’-NP Manipulation verbs ‘let’-NP-VP ‘have’-NP-VP ‘make’-NP ‘get’-(NP)

Direct (manip.) 5 9 31 6 17 4 2 4 5

Descriptive / / 4 / / / / / /

Direct (manip.)

Descriptive

2 1 3 8

/ / / /

 T. GivÓn

Perception-epistemic

Attract-attention

Descriptive

2 6 2 18 2

/ / / / /

utterance-epistemic

manipulative

descriptive

‘say’-WH/S ‘say’-dir. quote total:

/ / 127 (95.4%)

1 1 6

‘see’-(NP) ‘see’-NP-VP/S S, ‘see’ ‘look’-(at)-NP ‘look’. S

2D: Distribution of adult uses of modal patterns, Naomi-II Modality verbs

Direct (manip.)

Descriptive

7 5 2 2 1 5 3 1 2 7 / 1 / 1

/ / 1 / / 1 1 / / / 1 / 1 /

Direct (manip.)

Descriptive

9 / 2 2 1 4 /

/ 2 / / 1 / 1

Perception-epistemic

Direct (attention)

Descriptive

Direct (evidential)

‘see’-NP ‘see’-NP-VP (raising) ‘see’-WH/S S. ‘see’ ‘look’-at-NP

/ 1 1 / /

1 / / / 1

/ 1 1 1 /

‘will’-VP ‘can’-VP ‘could’-VP ‘should’-VP ‘may’-VP ‘gonna’-VP ‘have’-to-VP ‘gotta’-VP ‘want’-NP ‘want’-to-VP ‘like’-NP ‘need’-NP ‘try’-to-VP ‘come-(and)-V (serial) Manipulation verbs ‘let’-NP-VP ‘have’-NP ‘have’-NP-VP ‘want’-NP-VP ‘make’-NP ‘make’-NP-VP ‘get’-NP-VP

The ontogeny of complex verb phrases 

‘show’-NP (ellipsis) 1 ‘hear’-(ellipsis) / ‘listen’ (ellipsis) 2 Cognition-epistemic Direct (epist. quant.)

/ / / Descriptive

‘know’-(ellipsis) S, ‘know’ ‘know’-WH/S ‘know’-S ‘remember-(S/WH/S) ‘think’-S S, ‘guess’

5 1 3 1 5 3 1

/ / / / / / /

Utterance-epistemic

Direct (manip.)

Descriptive

1 2 /

/ 1 /

82 (+4) (87.7%)

12

‘say’-dir.quote ‘say’-NP ‘say’-WH/S total:

4

2E: Distribution of child uses of modal patterns, Nina-II Modality verbs ‘will’-VP ‘can’-VP ‘gonna’-VP ‘go-and-V’ (serial) ‘go’-LOC ‘go’-to-V ‘come-(and)-V (serial) ‘want’-NP ‘wanna’-VP ‘like’-NP

Direct (manip.)

Descriptive

11 1 6 2 3 3 1 27 25 /

1 / 1 / 2 / / 1 2 1

17 / 2 13 5

/ 14 / / 2

Direct (manip.)

Descriptive

2 1 1 1 1

2 / / / /

Manipulation verbs ‘let’-NP-VP ‘have’-NP ‘make’-NP ‘get’-(NP) ‘get’-LOC (incho.) Perception-epistemic ‘see’-(NP) ‘see’-S S. ‘see’ ‘look’-(at)-NP ‘show-DAT-NP

/ 1 /

 T. GivÓn

Cognition-epistemic ‘know’-WH/S ‘pretend’-S ‘wonder’-WH/S total:

Direct (epist. quant.)

Descriptive

Direct (manip.)

2 / 3 127 (+1) (83.1%)

/ / / 26

/ 1 / 1

2F: Distribution of adult uses of modal patterns in Nina-II Modality verbs

Direct (manip.)

Descriptive

14 1 16 2 13 / / 7 21 / 1 1 14 27 2 6 / / 1 2 1 1

6 / 8 / / 3 / 1 6 7 / / 1 2 / / 4 2 / / / /

Manipulation verbs

Direct (manip.)

Descriptive

‘let’-NP-VP ‘have’-NP ‘want’-NP-VP ‘would-like’-NP-VP ‘make’-NP ‘make’-NP-(into)-NP ‘get’-NP-(NP) ‘get’-LOC (incho.) ‘get’-to-VP (incho.)

12 / 5 1 8 2 6 2 1

/ 21 / / / / / 1 /

‘will’-VP ‘would’-VP ‘can’-VP ‘may’-VP ‘shall-VP ‘must’-VP ‘might’-VP ‘have-to’-VP ‘gonna’-VP ‘go’-LOC/WH ‘go’-(and)-V (serial) ‘come’-(and)-V (serial) ‘want’-NP/WH ‘want’-to-VP ‘would-like’-NP ‘would-like’-to-VP ‘like’-NP ‘like’-to-VP ‘need’-NP ‘be-ready’-(to-VP) ‘finish’-NP ‘why-don’t-you’-VP

The ontogeny of complex verb phrases 

Perception-epistemic

Direct (attention)

Descriptive

Direct (evidential)

14 3 2 1 1 1 2 / / 3

5 / / / / / / 1 1 /

3 / / / / / / / / /

Cognition-epistemic

Direct (epist. quant.)

Descriptive

Direct (attention)

‘know’-WH/S ‘remember’-NP ‘think’-S S, ‘think’ ‘think’-if-S ‘understand’-(ellipsis) S, ‘guess’ ‘bet’-S ‘pretend’-S Utterance-epistemic

2 / 8 1 1 / 1 1 1 Direct (manip.)

/ / / / / 1 / / / Descriptive

/ 2 / / / / / / /

‘tell’-WH/S ‘ask’-for-NP total:

1 / 210 (+5) (73.4%)

/ 1 71

‘see’-(NP) ‘see’-WH/S ‘see’-if-S ‘see’-S ‘look’-at-NP ‘look’-LOC ‘look’-WH/S ‘look-like’-NP ‘watch’-NP ‘show’-DAT-(NP)

5

APPENDIX 3: Distribution of direct speech-act vs. descriptive use of complex modal expressions by children and adults in stage-III transcripts 3A: Distribution of child uses of modal patterns, Eve-III Modality verbs

Direct (manip.)

Descriptive

‘will’ ‘can’-VP ‘gonna’-VP ‘go’-V (serial) ‘want’-NP ‘want’-to-VP ‘need’-NP ‘have-to-VP ‘like’-(ellipsis)

2 7 12 4 6 1 1 12 1

2 2 2 / / / / / /

 T. GivÓn

Manipulation verbs ‘let’-NP-VP ‘get’-NP ‘have’-NP ‘make’-NP Perception-epistemic ‘see’-NP ‘see’-S ‘see’-NP-VP ‘look’-NP.S Cognition-epistemic ‘think’, S ‘guess’-S ‘find-out’-WH/S ‘remember’-S total:

Direct (manip.)

Descriptive

6 4 (context) 4 (context) 9 (context)

/ / / /

Direct (attention)

Descriptive

2 2 (raising) 1

2 / 5 /

Direct (epist.quant.)

Descriptive

Direct (attention)

1 1 1 / 83 (+1) (91.3%)

/ / / / 8

/ / / 1 1

3B: Distribution of adult uses of modal patterns, Eve-III Modality verbs ‘will’-VP ‘would’-VP ‘can’-VP ‘could’-VP ‘may’-VP ‘might’-VP ‘shall’-VP ‘should’-VP ‘must’-VP ‘have-to’-VP ‘’ve-got-to’-VP ‘supposed-to’-VP ‘go’-LOC ‘gonna’-VP ‘want’-NP ‘want’-to-VP ‘would-like’-NP ‘like’-NP ‘need’-NP ‘try’-to-VP ‘go’-(and)-V (serial) ‘come’-(and)-V (serial)

Direct (manip.)

Descriptive

30 5 15 1 5 / 6 7 2 7 1 / 3 (context) 9 4 3 2 / 5 1 (context) 5 (context) 1 (context)

9 1 5 / / 1 / / / 1 / 1 1 2 / / / 1 / / / /

/

‘wait’-and-V (serial) ‘(had)-better’-VP ‘why-don’t-you’-VP Manipulation

The ontogeny of complex verb phrases 

1 (context) 2 5

/ / /

Verbs

Direct (manip.)

‘let’-NP-VP ‘want’-NP-VP ‘would-like-NP-to-VP ‘make’-NP ‘make’-NP-VP/PRED ‘have’-NP ‘take’-NP ‘get’-NP

3 1 1 10(context) 6 (context) 8 (context) 2 (context) 2 (context)

/ / / 2 / / / /

Perception-epistemic

Direct (attention)

Descriptive

‘see’-NP ‘see’-if-S ‘see’-NP-VP (raising) ‘look’-(at-NP) ‘watch’-NP-VP ‘hear’ ‘listen’ ‘feel’-ADV

6 1 / 3 1 1 1 /

Descriptive

3 / 1 / / / / /

Cognition-epistemic

Direct (epist. quant.)

Descriptive

Direct (attention)

‘know’-if-S ‘know’-S S, ‘know’ ‘find-out’-WH/S ‘remember’-(WH/S) ‘forget’-about-NP ‘think’-S ‘think’-(ellipsis) ‘guess’-S ‘be-sure’-(ellipsis)

1 2 1 / / / 11 1 2 1

/ / / / / 1 / / / /

/ / / 1 6 / / / / /

Utterance-epistemic

Manipulative

Descriptive

2 / 1 188 (+7) (83.5%)

/ 1 / 30

‘say’-dir.quote ‘tell’-WH/S ‘ask’-DAT total:

7

 T. GivÓn

3C: Distribution of child uses of modal patterns, Naomi-III Modality verbs

Direct (manip.)

Descriptive

10 1 7 4 36 4 21 3 1 / /

/ / 3 / / / / / / 6 2

Direct (manip.)

Descriptive)

1

/

/ 3 (icho.)

1 2 3

Direct (attention)

Descriptive

Direct (manip.)

2 1 1 2 2

/ / / / /

1 2 / / /

Cognition-epistemic

Direct (epist. quant.)

Descriptive

Direct (attention)

‘think’-WH utterance-epistemic ‘say’-WH total:

2 direct / 104 (+3) (85.2%)

/ descriptive 1 15

/

‘can’-VP ‘could’-VP ‘gonna’-VP ‘go’-LOC ‘want’-NP ‘wanna’-(ellipsis) ‘wanna’-VP ‘need’-NP ‘have’-to-VP ‘like’-NP ‘finish’-(ellipsis) Manipulation verbs ‘let’-NP-VP ‘get’-NP ‘have’-(NP) ‘get’-NP ‘get’-LOC Perception-epistemic ‘see’-NP ‘see’-(ellipsis) ‘see’-NP-VP (raising) ‘look’-(at)-NP ‘look’-at-NP-VP (raising)

3D: Distribution of adult uses of modal patterns, Naomi-III Modality verbs

Direct (anip.)

Descriptive

‘will’-VP ‘would’-VP ‘can’-VP ‘could’-VP ‘be-able’-to-VP ‘should’-VP

9 / 9 / / 1

3 4 4 2 1 /

3

‘have-to’-VP ‘seem-to-be’-VP ‘go’-LOC ‘gonna’-VP ‘go’-(and)-V (serial) ‘come’-(and)-V (serial) ‘want’-NP ‘want’-to-VP ‘need’-NP ‘need’-to-VP ‘would-like’-to-VP ‘have’-to-VP ‘like’-NP ‘finish’-(VP/NP) ‘why-don’t-you’-VP Manipulation verbs ‘let’-NP-VP ‘want’-NP-VP ‘wish’-S ‘ask’-NP(VP) ‘help’-DAT-(VP) ‘get’(DAT)-NP ‘ve-got’-NP ‘be-time-for’-NP-to-VP Perception-epistemic ‘see’-NP ‘see’-(ellipsis) ‘look’-(at-NP) ‘listen’, S Cognition-epistemic ‘know’-(ellipsis) ‘know’-WH/S ‘remember’-(WH/S) ‘think’-S ‘think’-(ellipsis) ‘wonder’-WH/S ‘mean’-dir.quote Utterance-epistemic ‘say’ ‘tell’ total:

The ontogeny of complex verb phrases 

1 / 3 (context) 6 1 (context) 1 (context) 9 9 3 1 1 3 / / 2

/ 1 1 19 / / / 1 / / / / 7 2 /

Direct (manip.)

Descriptive

3 7 1 / 2 5 (context) / 1

3 / / 1 2 / 2 /

Direct (attention)

Descriptive

Epist. quantifier

/ 1 2 1

3 1 / /

1 / / /

Direct (epist. quant.) 3 4 / 3 1 1 1 Direct (manip.) 4 1 98 (+3) (61.9%)

Descriptive / / / / / / /

Direct (attention) / / 2 / / / /

Descriptive 4 1 62

3

 T. GivÓn

3E: Distribution of child uses of modal patterns, Nina-III Modality verbs

Direct (manip.)

Descriptive

5 2 7 1 2 7 7 (context) 5 1 5 / 5 (context) 2

6 1 3 / 1 9 8 / / 1 / / /

Direct (manip.)

Descriptive

‘let’-NP-VP ‘have’-NP ‘make’-NP ‘make’-NP-VP (caus.) ‘get’-NP ‘get’-LOC (incho.)

20 / 11 (context) 1 1 (context) /

/ 5 / 1 / 3

Perception-epistemic

Direct (attention)

Descriptive

‘will’-VP ‘would’-VP ‘can’-VP ‘could’-VP ‘should’-VP ‘gonna’-VP ‘go’-LOC ‘want’-NP ‘wanna’-(ellipsis) ‘wanna’-VP ‘like’-NP ‘try’-VP ‘it’s time’-(for you-to-VP) Manipulation verbs

‘see’-(ellipsis) ‘see’-WH/S ‘see’-S S. ‘see’ ‘look’-at-NP ‘look’-at-NP-VP (raising) Cognition-epistemic ‘know’-NP ‘forget’-(ellipses) ‘forget’-to-VP ‘understand’-(NP)–(ellipsis) Utterance-epistemic ‘say’-dir. quote ‘say’-S total:

3 1 2 1 5

/ / / / /

Direct (epist. quant.)

Descriptive

1 / / /

/ 1 1 2

Direct (manip.)

Descriptive

/ / 95(66.4%)

5 1 48

The ontogeny of complex verb phrases 

3F: Distribution of adult uses of modal patterns, Nina-III Modality verbs ‘will’-VP ‘can’-VP ‘could’-VP ‘shall’-VP ‘should’-VP ‘must’-VP ‘have-to’-VP ‘go’-LOC ‘gonna’-VP ‘go-(and)-V (serial) ‘go’-to-V ‘come’-(ellipsis) ‘want’-NP/WH ‘want’-to-VP ‘love’-to-VP Modality (cont.) ‘would-like’-to-VP ‘like’-NP ‘like’-to-VP ‘need’-NP/WH ‘be-ready’-to-VP ‘try’-NP ‘try’-to-VP ‘be-time’-(for-NP)-to-VP ‘be-better’-VP Manipulation verbs ‘let’-NP-VP ‘have’-NP ‘make’-NP/WH ‘make-NP-VP ‘get’-NP-ADJ ‘take’-NP-ADJ Perception-epistemic ‘see’-NP ‘see’-(ellipsis) ‘see’-WH/S ‘see’-S ‘see’, S S, ‘see’ ‘see’-if-S

Direct (manip.)

Descriptive

4 18 / 16 2 / 5 3 1 2 1 / 2 23 /

6 6 1 / 3 1 3 14 22 / 1 1 / 2 1

Direct (manip.)

Descriptive

5 / / / / 2 1 1 3

/ 2 4 3 2 / 1 / /

Direct (manip.)

Descriptive

8 / 3 (context) 7 / /

/ 4 2 4 2 1

Direct (attention)

Descriptive

4 11 2 2 1 1 1

4 / / / / / /

 T. GivÓn

‘look’-(at-NP) ‘look’-at-NP-VP (raising) ‘look’-WH/S WH/S, ‘look’ ‘look’-for-NP ‘look’-ADJ ‘watch’-NP ‘feel’-WH/S ‘feel’-ADJ

15 1 4 1 / / / 1 /

/ / / / 2 1 1 / 1

Direct (epist. quant.)

Descriptive

Direct (attention)

10 4 3 / 3 22 2 2 2 / 3 2 1 /

/ 1 / / / / / / / 1 / / / 1

/ / / 1 / / / / / / / / / /

Utterance-epistemic

Direct (manip.)

Descriptive

Direct (evidential)

‘say’-dir.quote ‘say’-WH S, ‘say’ ‘tell’-S total:

1 / / / 206 (+3) (67.8%)

/ 1 / / 99

/ / 1 1 3

Cognition-epistemic ‘know’-(ellipsis) ‘know’-WH/S WH/S, ‘know’ ‘remember’-NP ‘wonder’-WH/S ‘think’-S ‘think’-(ellipsis) WH/S-‘think’ ‘guess’(-S) ‘understand’-WH ‘mean’-S S, ‘mean’ ‘pretend’-to-VP ‘forget’-to-VP

References Bates, E. 1976. Language in Context: The Acquisition of Pragmatics. New York NY: Academic Press. Bates, E., Camioni, L. & Volterra, V. 1975. The acquisition of performatives prior to speech. Merril-Palmer Quarterly 21. Also in E. Ochs & B. Schieffelin (Eds), 1979. Boye, K. & Harder, P. 2007. Complement-taking predicates: Usage and linguistic structure. Studies in Language 31(3): 3. Carter, A. 1974. Communication in the Sensory-Motor Period. Ph.D. dissertation, UCSB. Chafe, W. 1994. Discourse, Consciousness and Time: The Flow and Displacement of Conscious Experience in Speaking and Writing. Chicago IL: University of Chicago Press. Chafe, W. 1997. Polyphonic topic development. In T. Givón (Ed.), 1997.

The ontogeny of complex verb phrases 

Cheney, D. & Seyfarth, R. 2007. Baboon Metaphysics. Chicago IL: University of Chicago Press. Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: The MIT Press. Diessel. H. 2005. The Acquisition of Complex Sentences, Cambridge: CUP. Diessel, H. & Tomasello, M. 2001. The acquisition of finite complement clauses in English: A corpus-based analysis. Cognitive Linguistics 12(2). Ervin-Tripp, S. 1970. Discourse agreement: How children answer questions. In J.R. Hayes (Ed.), Cognition and the Development of Language. New York NY: Wiley. Ervin-Tripp, S. & Küntay, A. 1997. On occasioning and structure of conversational stories. In T. Givón (Ed.), 1997. Givón, T. 1979. On Understanding Grammar. New York NY: Academic Press Givón, T. 1993. English Grammar: A Function-Based Introduction, 2 Vols. Amsterdam: John Benjamins. Givón, T. (Ed.). 1997. Conversation: Cognitive, Communicative and Social Perspectives [Typological Studies in Language 34]. Amsterdam: John Benjamins. Givón, T. 2001. Syntax: An Introduction, 2 Vols. Amsterdam: John Benjamins. Givón, T. 2008. The Genesis of Syntactic Complexity: Diachrony, Ontogeny, Neuro-Cognition, Evolution. Amsterdam: John Benjamins Heine, B., Claudi, U. & Hünnemeyer, F. 1991. Grammaticalization: A Conceptual Framework. Chicago IL: University of Chicago Press Heine, B. & Kuteva, T. 2007. The Genesis of Grammar: A Reconstruction. Oxford: OUP. Hopper, P. & Traugott, E.C. 1993. Grammaticalization. Cambridge: CUP. Linell, P. & Karolija, N. 1997. Coherence in multi-party conversation: Episodes and context in interaction. In T. Givón (Ed.), 1997. MacWhinney, B. & Snow, C. 1985. The Child Language Data Exchange System. J. of Child Language 17. Meltzoff, A.N. 1999. Origins of theory of mind, cognition and communication. J. of Communication Disorders 32. Meltzoff, A.N. 2000. Imitation as a mechanism of social cognition: Origins of empathy, theory of mind and the representation of action. In Blackwell Handbook of Child Cognitive Development, U. Goswami (Ed.). Oxford: Blackwell. Newmeyer, F. 2008. Complementing Boye and Harder on complements: What conversational English tells us about the nature of grammar. Ms, University of Washington. Ochs, E., Schieffelin, B. & Platt, M. 1979. Propositions across utterances and speakers. In E. Ochs & B. Schieffelin (Eds), 1979. Ochs, E. & Schieffelin, B. (Eds). 1979. Developmental Pragmatics. New York NY: Academic Press Pearson, T.R. 1985. A Short History of a Small Place. New York NY: Ballantine. Povinelli, D.J. & deBlois, S. 1992. Young children’s understanding of knowledge formation in themselves and others. J. of Comparative Psychology 106. Scollon, R. 1976. Conversations with a One-Year-Old Child. Honolulu HI: University of Hawaii Press Thompson, S. 2002. Object complements and conversation. Studies in Language 26. Thompson, S.A. & Mulac, A. 1991. A quantitative perspective on the grammaticalization of epistemic parentheticals in English. In E.C. Traugott & B. Heine (Eds), 1991. Traugott, E. & Heine, B. (Eds). 1991. Approaches to Grammaticalization, TSL 19, Amsterdam: J. Benjamins. Wellman, H.M. 1990 Children’s Theories of Mind. Cambridge MA: The MIT Press

part iii

Cognition and neurology

Syntactic complexity versus concatenation in a verbal production task Marjorie Barker & Eric Pederson University of Oregon

We tested whether the speaker’s communicative intent drives the selection of grammatical constructions. Participants viewed complex human action video stimuli and were asked to respond in detail to a single question for each video concerning either what had happened (eliciting descriptions) or why a particular event had occurred (eliciting explanations). We predicted that responses to the why questions would contain more syntactically complex constructions (specifically verbal complements), while responses to the what questions would be more concatenated. The experimental results with these stimuli did not uphold the first part of the hypothesis: complexity in the form of syntactic embedding was statistically equivalent under both conditions. However, there was significantly more concatenation in the form of coordination in the what condition.

1. Introduction The ontogeny of syntactic complexity has been discussed from evolutionary, developmental, and diachronic points of view. These various perspectives have both complemented and challenged each other, resulting in a multifaceted understanding of the nature and history of complex syntax. Some degree of syntactic complexity can be seen as the natural consequence of the evolution of a rich communication system, and syntactic complexity has gained particular prominence in literature on the evolution of language. The issue of syntactic recursion in particular has generated considerable debate (see Hauser et al. 2002; Everett 2005; Jackendoff & Pinker 2005; and Bickerton, this volume). Generally speaking, evolutionary accounts must offer explanations of the development of syntactic complexity in terms of adaptive benefit: slight processing gains can deliver a substantial advantage for quick communication and decision-making and consequently for survival. In the absence of early Modern Human language data, discussion remains largely speculative; however, comparing accounts of language data from a variety of contexts,

 Marjorie Barker & Eric Pederson

Wray & Grace (2007) argue that much of what is considered syntactic complexity arose from the interaction of groups of intimates, where having formulaic systems to handle routine communication minimized the processing effort involved in spoken interactions. Developmental descriptions of child language acquisition follow the growth of a contemporary human linguistic system, sometimes starting from birth, and try to explain how verbs representing separate but frequently-connected events may move through stages of paratactic association (coordination) to syntactic complexity (subordination) to complex verbal constructions. Much child language research on the development of complex constructions deals with relative clauses (e.g., Diessel & Rojas Nieto, both in this volume), where it is possible to compare expressions of the same meaning using either a coordinated or subordinated structure. Givón (also this volume) outlines a step-by-step progression for complement constructions in early child language. In this account, the child’s first verb complements are produced in cooperation with the interlocutor; the child provides the ‘complement’ and semantic core of the expression, while the adult models the ‘main’ clause showing the epistemic or deontic perspective. This particular study dovetails nicely with Wray & Grace’s account in that the intimate setting gives rise to more complex constructions. Our understanding remains limited however, since at a given stage for a child, we cannot know what drives any particular sentence to be produced as more or less syntactically complex. Similarly, grammaticalization theory accounts for the various steps in the process of morphosyntactic change from concatenated structures to embedded structures; such diachronic accounts give insight into what kinds of structural change are possible. However, these are descriptions of examples following a potentially universal grammaticalization path. We lack the predictive power to say why certain constructions in some languages follow the predicted path while similar constructions in other languages do not. In essence, we lack controlled language data showing how individual speakers make constructional decisions online. Clearly, certain genres typically demonstrate greater syntactic complexity. For example, formal written texts tend to have far more subordination than spontaneous speech, which relies more heavily on concatenated structures (e.g., Chafe 1985; Givón 1998). Written language reflects editing processes which are less subject to immediate processing constraints, so written data masks and compresses the online decisions that ultimately yield the linguistic form that is produced. It is evident that each of these approaches has inherent limitations for discovering the factors that lead to complexity. Lacking from both evolution and diachrony is any data on how language is used online. Explanations must remain hypothetical since we cannot gather data on controlled language production and comprehension outside of the laboratory. Most studies of child language development also examine essentially random, uncontrolled language production. Within a single spoken genre, there has been a dearth of empirical study examining the

Syntactic complexity vs. concatenation in a verbal production task 

conditions under which greater or lesser syntactic complexity may be found. Tomlin (1985) examined the production in oral and written narrative of subordinated clauses, but in addition to embedded clauses the study included participial and infinitive clauses, whether embedded or not, clauses introduced by subordinating conjunctions such as while and although, and adverbial clauses. He showed that subordinated clauses in oral and written narrative are reliably associated with background information. However, embedded complements tended to be exceptions to this finding. Attempting to explain how complexity varies in extemporaneous speech will help to build a model of the origin of complexity in general, since the synchronic choices behind evolutionary and diachronic change must ultimately derive from immediate expressive demands. Examining how complexity varies online today will help to build a model of the origin of complexity in general. Accordingly, the task in this chapter is to explore spoken modern language production in controlled environments. Under what conditions will we find speakers using more concatenated structures and under what conditions will they use more syntactically complex constructions?

1.1 Motivations for complexity in oral narrative Our experiment was designed to elicit complex syntax in speech. As a measure of syntactic complexity, we focused on complement clauses. Givón (2001) describes three functional domains for complement clauses in human language, namely, to express perception-cognition-utterance, modality, and manipulation. Modal complements involve desires (want, need), cognition involves beliefs (think, know) and manipulation involves causation (make, force). All of these verbs are used to explain why certain actions and behaviors take place, and a pervasive feature of these domains is evaluation. When speakers talk about perception and cognition, they evaluate mental activity, whether their own or that of other people, and express a basis for confidence in the subordinated event. For manipulation and modality, speakers use the main verb to explain the motivation for the action in the subordinated event. We might further propose that for English, information about manipulation and modality is most readily expressed using subordination rather than concatenation (for example, ‘he wanted to find the missing snake’ would be awkward to express in English using two coordinated clauses). To some extent, the nature of the events being described may influence the complexity of the description. However, a complex series of events may be described in any number of ways depending on the communicative goals of the speaker. We hypothesized that when speakers are asked to talk about events, they will use more complement constructions when asked to give reasons and motivations for these events. In

 Marjorie Barker & Eric Pederson

other words, event descriptions which express causal and other explanatory relationships will more naturally tend toward a preponderance of embedded structures which tightly bind sub-events to one another. A second hypothesis was conversely related to the first: that event descriptions which express details of the events will tend to have a greater preponderance of concatenated structures because these allow the speaker to package more details about events together. To manipulate the communicative goals of the speakers, we presented visual stimuli in the form of video clips and asked experimental participants to answer questions about either what had happened in the video (to encourage description of details) or about why the events had happened (to encourage explanation of reasons and motivation).

2. Methodology We investigated the conditions under which people use complex syntax (i.e., the hierarchical structures of complementation or relativization) as opposed to simple syntax (i.e., the conjoined structures of coordination, or separate clauses) to describe associated events. Though our hypothesis specifically predicted an increase in complement clauses, we also counted relative clause use, since these are also embedded. Based on a pilot study, we determined that the most effective way to cue participants to describe or explain the videos was to ask them questions which prompted for either detailed descriptions (what) or explanations (why), rather than asking them to either ‘describe’ or ‘explain’ the events that they had seen. We hypothesized that asking questions about why a particular action was taken would point participants toward explanation, because speakers would talk about motivations and causality, requiring a relatively greater use of complement-taking verbs and a relatively smaller number of concatenations. Conversely, we predicted that in response to what questions, speakers would make greater use of concatenation, but would use relatively fewer complement structures, since speakers would be talking about concrete visible actions and presumably not about the reasons the events took place. Eight experimental and one training video clips, each approximately one minute in length, were taken from separate episodes of the USA Network TV series Monk. Segments were chosen that depicted a sequence of events telling a brief coherent story in which two or three main characters were involved. Participants were randomly assigned to either of the two conditions (why vs. what questions) and were unaware of the two groups. After viewing each video clip they were asked a question about the video and their response was audio recorded. Video clips were presented in a fixed order. All participants in the why condition were asked only why questions and all participants in the what condition were

Syntactic complexity vs. concatenation in a verbal production task 

asked only what questions. (Brief descriptions of the video clips and the why/what questions asked about the videos are included in Appendix A.)

2.1 Why condition Fifteen experimental participants (7 female, 8 male; median age 21) were recruited from the University of Oregon Psychology & Linguistics Human Subjects pool. Participants first watched a sample video clip, then were given the following instructions: You will be watching eight more video clips selected from the same TV series. After you watch each one, I will ask you a question about why something happened in the video. Please give a full and complete explanation in answer to the question, in other words, several sentences, connections between various events, rather than simply the immediate cause. For example, for the sample video, if the question was “Why does Monk use a handkerchief to hold onto Kevin’s pants?” you wouldn’t want to say it’s just because Monk has OCD. You need to talk about the multiple things that are going on, such as Kevin’s need to look into the house the need not to touch the floor the need for a counterweight Monk’s fear of germs etc.

2.2 What condition Fifteen experimental participants (9 female, 6 male; median age 19) were recruited from the University of Oregon Psychology & Linguistics Human Subjects pool. Participants first watched a sample video clip, then were given the following instructions: You will be watching eight more video clips selected from the same TV series. After you watch each one, I will ask you a question about what happened in the video. Please give a full and complete description in answer to the question, in other words, several sentences rather than simply an undetailed response. For example, for the sample video, if the question was “What does Kevin do in the process of getting the pencil?” you wouldn’t want to say just that he gets it with the shovel. You need to talk about the multiple things that you saw, such as Kevin’s having Monk hold his pocket his leaning into the house reaching for the chair getting the shovel using the shovel to scoop up the pencil etc.

 Marjorie Barker & Eric Pederson

Participants viewed one video clip at a time, then were asked one question about the video they had just viewed, and audio-recorded as they produced an oral response to the question.

3. Results For each speech passage produced, we determined the length of the response in seconds. We also counted the number of transcribed lines for each response as a secondary comparative measure in case rate of speech differed greatly between participants. For each response, coordinate, complement, and relative clauses were marked and counted. (Sample responses to the two questions associated with one experimental video clip are given in Appendix B.) Speech passages were divided into phrasal units as the basic processing element. We considered these to be an indication of planned speech units, hence mental association, in addition to and possibly more reliable than lexical coordinators like and, but, and or, which are often used as discourse markers to show narrative continuation rather than syntactic connection. In other words, without knowing the intonational packaging it is impossible to tell from transcripts whether and marks coordinate clauses, since coordination joins syntactically complete expressions. Complementation constructions, on the other hand, can span across intonation contours, because the first part is clearly incomplete and awaits the completion in the second. There may well be long-term planning across boundaries, but this is difficult to measure and not necessarily consistent. Therefore, clausal coordination was described as clauses connected by and, but, or, or Ø. but verbs were counted as coordinated only if they fell under the same intonation contour. Complementation was determined syntactically, in other words even when the complement construction occurred across a phrasal boundary — though this was quite rare (6 out of 591 total complements). Relativization was also determined syntactically (spanning phrasal boundaries, 5 out of 186 relative constructions). We did not count clauses from meta-cognition or commentary — i.e., “I guess he was having lunch” does not count as complex for purposes of describing/explaining the action of the video. “Monk thought there was another snake” does count as complex, because it describes what is going on in the video. The verb like used for quotation was counted as cognition/utterance complementation. Fifteen participants who answered one why question after viewing each video produced eight narratives each for a total of 120 explanations. Fifteen participants who answered one what question after viewing each video also produced eight narratives each for a total of 120 descriptions. Results are shown in Table 1.

Syntactic complexity vs. concatenation in a verbal production task 

Table 1. Comparison of what and why conditions.

What condition (n=120)

Why condition (n=120)

Response length Seconds Lines Associated clauses/response Relative Complement Coordinated Associated clauses/second Relative Complement Coordinated

Mean (SD)

Mean (SD)

24.4 (12.1) 3.8 (1.85)

23.1 (7.93) 4.0 (1.43)

0.75 (1.00) 2.54 (2.21) 1.32 (1.22)

0.80 (0.94) 2.38 (1.56) 0.72 (0.85)

.031 (.042) .103 (.083) .055 (.048)

.035 (.044) .106 (.065) .029 (.035)

ANOVAs were run for all three measures comparing why vs. what responses. These analytic measures showed no statistically significant difference in the rate of complementation (single factor ANOVA p=.8) nor relativization (single factor ANOVA p=.5) between the why and the what conditions. However, coordination was significantly greater in the what condition (single factor ANOVA p<.001).

3.1 Complement types We further analyzed complement clauses by classifying them into six groups: 1. Speech — describing speech or semiotics (she yells at Monk to help, he pointed out that he saw, she’s saying ‘he’s gonna get me!’). 2. Cognition — Describing mental activity (he knew how to do it, he couldn’t figure out how to move it). 3. Manipulation — describing the use of someone or something to accomplish a task (he got the ferris wheel to move, he used his legs to press the door shut). 4. Modality — describing attempt, intent, obligation, ability, or possibility (he tried to climb on, he wants to know). 5. Aspect — describing inception, termination, continuation, success, or failure (it ends up breaking, he started to back out). 6. Other — where the subordinated clause describes intent (he climbed up to prevent the guy from getting her, he reached to grab the cloth). Although the overall frequency of complements was not significantly different between conditions, the types of complements used by speakers varied noticeably between the what and why conditions. See Figure 1.

 Marjorie Barker & Eric Pederson 50% 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% ch

tio n

ee sp

ul ip an

m

co gn i

at io n

r ot he

ct pe as

m

od a

lit y

WHAT Condition WHY Condition

Figure 1. Comparison of complementation types.

4. Discussion The hypothesis that people will use more complex syntax when asked to explain an event than when simply describing was not supported, and thus must be rejected for tasks such as this experiment. The rate of complementation in online speech appears, at least in these tasks, to be fairly consistent. However, we found that the type of complement verbs used does vary with expressive purposes. Participants answering why questions were far more likely to use complements of cognition than those answering what questions. There was no cognitive loading beyond the task of giving descriptions or explanations; participants were just talking, concentrating only on what they said. Reporting observed events is a common, low-effort task. Based on these results, we might propose that the rate of complementation is stable in speech, possibly because of cognitive constraints on how much embedded information can be processed in a given unit of time. In other words, people typically embed at a rate near their current production capacity. It may well be that in (frequent) situations where the language production system is sharing cognitive resources with other processes that greater differentiation of complexity would emerge. While the rate of production of complements and relative clauses was stable across conditions, the amount of coordination was significantly higher in the what condition. This increase of coordination effectively means that there were more clauses within the phrasal unit. Multi-clausal intonation contours in the what condition could be interpreted as reflecting a conceptual grouping of events. Put another way, this result suggests there is a difference in chunking behavior across the two conditions. The motivation to describe fosters a strategy of concatenated packaging.

Syntactic complexity vs. concatenation in a verbal production task 

However, we do not see a difference in embedding — i.e., complement and/or relative clauses. If we take embedding, but not concatenation, as indicative of syntactic complexity, then this is consistent with the suggestion by Fernandez-Duque in this volume that syntactic complexity may not be functionally related to the general cognitive processes of chunking. In summary, then, our main hypothesis was not upheld; participants did not produce more syntactically complex structures in extemporaneous speech when asked why questions that required them to provide explanations of events they had viewed than when asked what questions that required them to simply describe the same events. A related but secondary hypothesis was confirmed, however; participants produced significantly more concatenated clauses when answering what questions than when answering why questions. Several observations may be made in connection with these results. Manipulation of speaker intent did produce a change in linguistic behavior. Participants treated the two conditions differently: they produced different verbs in response to why questions (explaining) than in response to what questions (describing), and they used much more coordination when describing than when explaining. Thus the methodology is useful for eliciting condition-dependent language. Further, we had predicted that speakers would employ more concatenation when responding to a what question. We thought that speakers responding to why questions would employ more hierarchical syntactic structure in relating these events, while responses to what questions would give their description of these same events a more flat syntactic structure. Instead, participants produced complements and relative clauses at an equal rate under both conditions, and those responding to what questions added an additional level of association using intonation. The stable rate of embedding (complementation and relative clauses) may indicate a ceiling effect, where speakers can handle only so much syntactic complexity in a given unit of time. However, the usage of verbal complements at least in English is highly subject to pragmatic considerations. Complement taking verbs are often employed in the service of politeness and perspective-taking. While such use can also be classified under speaker intent, this use is not as straightforward as with a simple description or explanation. The choice of whether or not to explicitly say ‘I thought’ or ‘he thought,’ for example, can relate to the impression the speaker wants to convey about the person performing the action. In everyday speech it is not unusual for speakers to omit the complement verb that marks belief, probably because it is evident from context whether an actor believes something. For example in the sentence, “Monk climbed up on the ferris wheel because the woman was in danger” we assume his belief in her danger was self-evident. In fact, to say that he climbed up ‘because he thought she was in danger’ conveys a sense that the speaker does not share the actor’s belief (cf. Malle et al. 2000; Barker & Givón 2005).

 Marjorie Barker & Eric Pederson

It is evident that linguistic complexity is multifaceted. While the focus of this study and this volume is syntactic complexity, there are in fact many features that contribute to complexity in language — processing load, informational content, and salience (situational, personal and interpersonal), for example. We have sought to explore motivational factors in the production of subordination and concatenation; a complete explanation of the phenomenon will involve multiple factors.

Appendix A: Description of stimuli, why and what questions Training video: Leaning — Monk and another man stand in the doorway of a house. They open the door and nobody is home. The other man asks Monk to hold the pocket of his pants and act as a counterweight so he can lean into the house and look around without touching the floor. He can’t reach a pencil on the sofa, so grabs a shovel outside the door and uses it to get the pencil. Monk pulls him back. Trial videos: Ball — Monk and a woman are in the office of a sports agent, who is talking on the phone. Monk takes out a tissue and wipes a basketball in a display case. The agent gets off the phone quickly and tells Monk to stop, explaining the “stain” on the ball is Michael Jordan’s autograph. Monk proposes that Michael could sign the ball again and the agent sarcastically agrees; Monk says he is relieved and the woman grimaces. Why does the woman look disgusted when Monk says he is so relieved? What does Monk do with the tissue? Car — Two men hand Monk the keys to a car parked at a curb. Monk takes the keys and goes to the driver’s side of the car, but climbs first into the back seat of the car, then into the driver’s seat, while the woman with him gets in the passenger seat. He maneuvers with difficulty out of the parking space while other drivers honk. There is a crash; the car has crashed into a light pole. Monk tells the woman that she told him to turn, so he turned. Why did Monk crash into the light pole at the end? What did Monk do with the car? Dog — Monk is running through the aisles of a store that is closed. A Doberman is chasing him. He goes into a customer service booth, lies on the floor, and holds the swinging door closed with his feet. He looks around, grabs a phone, and dials. The scene cuts to a restaurant where a man and woman are eating. The woman’s phone rings; she answers but there is no response, only barking. The man listens and then says “let’s go.” They hurriedly leave the restaurant. Why did the man and woman leave the restaurant in a hurry? What did Monk do to escape the dog? Ferris wheel — A woman is in the seat of a ferris wheel; a man is climbing toward her on the bars of the ride. Monk is at the controls of the ferris wheel, pushing buttons. The woman

Syntactic complexity vs. concatenation in a verbal production task 

shouts at Monk to get her down. The wheel starts moving, Monk pulls a lever out of the controls, and finally jumps up onto the ferris wheel. Why does Monk jump onto the ferris wheel? What does Monk do at the ferris wheel? Roof — Monk and another man look out an upper-story window; Monk points out a red rag on the chimney of the building. The other man climbs out onto the roof with Monk cautioning him to be careful. The man gets the rag, and Monk asks him to look for footprints, which he sees, and then climbs back to the room where he says that the rag is evidence. Why are Monk and Manny looking for footprints on the roof? What did Manny do on the roof? Room — Monk is in a nursing home room. He examines the furniture, walls, etc and questions a nurse about the occupant of the room and whether anything has been moved. A police officer is talking on a cell phone in the background describing what Monk is doing. He asks Monk to give an opinion, and Monk states that the man was murdered. The police officer seems surprised. Why was the captain surprised that Monk said the old man was murdered? What did Monk look at while investigating the old man’s room? Snake — Monk and a police officer are inside a house. The officer is putting a snake back into a cage while Monk acts panicky. When the snake is back and the lid closed, Monk examines the snakes, saying there is a feeding schedule for Curly, Larry, and Moe. He looks closely, shouts, then climbs onto the table. The officer runs into the room and asks what he’s doing. Monk replies that there are only two snakes in the cage; one must be loose. Why did Monk jump up on the table? What did the captain do with the snakes? Tie — A woman is talking to a man on a park bench, while Monk leans over them. The man says they have no proof; Monk stands up and holding his tie, speaks toward it, saying “come and get him.” There is no response. Monk runs toward a white van parked nearby and opens the back doors. He asks the policemen inside why they didn’t arrest the man; they respond that they heard nothing, and ask what happened to a stain on his tie. He responds that he finally got it out; they look dismayed. Why were the agents inside the van upset? What did you see Monk do with his tie?

Appendix B: Sample responses to one video clip (‘car’) What condition: What did Monk do with the car? B16 — Alright, it starts out with Monk taking, a pair of keys from a man, on a sidewalk, / and then he walks around the front of the car, / and he sits in the back seat, / and

 Marjorie Barker & Eric Pederson

then closes it and opens the door again I believe, and, goes to the front seat where he’s driving, / and he sticks his head out the window, / and, tries to back up and hits a car and it makes a noise, / but he drives away anyway and sticks his hand out the window to kinda wave people away, / they’re honking, and then, / uh, he hits, uh, I think it’s a light post, on, the right side of the car. / And gets out and is very frustrated. (38 sec) B17 — Monk got in the wrong door of the car, in the back seat, instead of the front seat; / he, then, reversed into another car, / and then kept going, / and the alarm was going off, / then he, swerved in, to the middle of the street, while another car was coming, and they honked at him, / and then he kept going, / and then, ran into, a pole. (25 sec) B18 — Monk got in the back seat first cause he was distracted, / and then he got into the driver’s seat, / and he, backed up, / and, since he was nervous, it seems like he, wasn’t paying attention and he backed up into a car behind him, / and then, continued to drive off, and cut off traffic, / and then, he drove into a pole. (24 sec) B19 — So, Monk got the keys from the guys standing on the curb, / and first he went to the back seat instead of getting into the driver’s seat, / and completely closed the door and then got into the driver’s seat, / and, s- proceeded to try to pull into traffic, it took him a few minutes, / he looked backwards, / and it sounded like he kinda backed into a car but I’m not really sure if he actually did, / and then he pulled out into traffic and cut somebody off because they honked the horn, / and then, he, ended up crashing into a light pole. (32 sec)

Why condition: Why did Monk crash into the light pole at the end? B1 — It appears that Monk, does not know how to drive, or he wouldn’t have gotten into the back seat of the car to begin with, / and in pulling out, he was very jerky, and, as he said to the woman, um, that she had told him to make the turn. / And so he followed her directions. (21 sec) B2 — Monk crashed into the light pole because he was so frazzled from what was going on, / he was trying to play it really cool, / um, getting out of that parking spot, cause he thought it was a really good spot, / and, um, he got stressed out, / and, that girl told him to turn, and supposedly he turned, / and, he was just listening to other people I guess. (23 sec) B3 — He crashed into the light pole because, uh, he was listening to the lady’s instructions, / and it didn’t look like he had, ever like driven a car before, / because, first when he gets into the wrong seat and, gets in the back seat of the car, instead of the driver’s seat; / so, uh, it looked like he had little experience of driving cars, / so, uh, he didn’t know when to turn right. (23 sec)

Syntactic complexity vs. concatenation in a verbal production task 

B4 — Uh Monk crashes into the light pole — light, pole, at the end, because he, uh doesn’t seem to be able to drive very well or be very sure of himself. (12 sec)

References Barker, M. & Givón, T. 2005. Representation of the interlocutor’s mind during conversation. In Other Minds: How Humans Bridge the Divide between Self and Others, B. Malle & S. Hodges (Eds), 223–238. New York NY: The Guilford Press. Bickerton, D. (in this volume) Recursion: Core of complexity or artifact of analysis. Chafe, W.L. 1985. Linguistic differences produced by differences between speaking and writing. In Literacy, Language, and Learning, N.T. David, R. Olson & A. Hildyard (Eds), 105–123. Cambridge: CUP. Diessel, H. (in this volume) On the role of frequency and similarity in the acquisition of subject and non-subject relative clauses. Everett, D. 2005. Cultural constraints on grammar and cognition in Pirahã. Current Anthropology 46: 621–641. Givón, T. 1998. Literacy and grammar. Institute of Cognitive and Decision Sciences Technical report No. 97–09. Givón, T. 2001. Syntax: An Introduction. Amsterdam: John Benjamins. Hauser, M., Chomsky, N. & Fitch, W.T. 2002. The faculty of language: What is it, who has it, and how did it evolve. Science 198: 1569–79. Jackendoff, R. & Pinker, S. 2005. The nature of the language faculty and its implications for evolution of language. Cognition 97: 211–25. Malle, B., Knobe, J., O’Laughlin, M.J., Pearce, G.E. & Nelson, S.E. 2000. Conceptual structure and social functions of behavior explanations: Beyond person – situation attributions. Journal of Personality and Social Psychology 79(3): 309–326. Rojas-Nieto, C. (in this volume) ‘Starting small’ effects in the acquisition of early relative constructions in Spanish. Tomlin, R.S. 1985. Foreground-background information and the syntax of subordination. Text 5: 85–122. Wray, A. & Grace, G.W. 2007. The consequences of talking to strangers: Evolutionary corollaries of socio-cultural influences on linguistic form. Lingua 117: 543–78.

The emergence of linguistic complexity Brian MacWhinney

Carnegie Mellon University Linguists have often argued that recursion produces linguistic complexity. However, recursion relies on preexisting processes such as lexical insertion, lexical combination, memory stacks, and methods of interpretation. In the brain, recursion is an emergent property of a set of adaptations that involve at least six processing systems. Linguistic complexity arises from the interplay of all six of these systems. The complexity of this neuronal support means that the full complexity of human language could not have arisen fortuitously at some single moment in evolution. However, there is evidence that some pieces of the six systems supporting complexity have developed more recently than others.

Human language is enormously complex. Each of us knows tens of thousands of words, and each of these words can specify a microcosm of meaning. When we put these words together into sentences and discourse, still further complexities arise in the form of collocations, phrases, and grammatical relations. Of the many forces that generate all this complexity, the one that has received the greatest attention is the process of recursion. Beginning with a seminal article by Bar-Hillel (1953), linguists have argued that recursion allows human beings to produce an infinite variety of possible sentences. Extending this vision, Chomsky and others have claimed that what makes human language distinct from the communication systems of all other animals is the fact that only humans have access to recursion. For decades, the characterization of recursion and its effects remained a task for linguists or computer scientists, with little connection to issues in biology. However, recently, Hauser, Chomsky, & Fitch (2002) (henceforth HC&F) hypothesized a link between recursion and human evolution. They distinguished between the faculty of language broadly defined (FLB) and the faculty of language narrowly defined (FLN). Within the FLB, there are many properties that human language shares with other communication systems. For example, both birds and humans rely on a vocal apparatus to produce a broadcast transmission of sound waves that is then decoded by an auditory apparatus. In both birds and humans, all members of the species possess the mechanisms for both production and perception. More generally, even outside the FLB, humans and other mammals share complex systems of social organization, memory systems, and methods for learning.

 Brian MacWhinney

For HC&F, none of these shared mechanisms are central to the faculty of human language, narrowly defined. They must be present for language to operate, but what is at the core of human language is recursion. This analysis places a heavy conceptual weight on the process of recursion. Given its centrality, it is surprising to find that HC&F fail to provide a clear account of how recursion operates linguistically.1 Instead, they simply note that recursion “takes a finite set of elements and yields a potentially infinite arrangement of discrete expressions.” In other words, recursion is whatever it takes to produce infinite generativity. But, if we are to understand recursion as a neurolinguistic process,2 we need to find out not just what it produces, but also how it operates. It is easy enough to provide some clear cases of the syntactic effects of recursion. As Bickerton (this volume), Shibatani (this volume), and others have noted, sentences with relative clauses provide just such clear examples. In the initial version of transformational grammar (Chomsky, 1957), a sentence like (3) was produced by a transformation that combined (1) and (2).

(1) The boy has a black jacket.

(2) The boy stole the bike.

(3) The boy who stole the bike has a black jacket.

However, in order to constrain the global power of transformations, Chomsky (1965) shifted away from a transformational view of recursion to a theory that generated recursion directly inside phrase structure rules, using the notion of a generalized phrase marker. In the case of example sentence (3), the insertion of the relative clause was then handled through rules like (4) through (6).

(4) S Æ NP + VP

(5) VP Æ V + NP

(6) NP Æ (D) N′ (S′)

It is the presence of the symbol “S” on the right-hand side of rule (6) that produces the insertion of a relative clause as the modifier of “the boy”. This process of 1. Tomalin (2007) shows how recursion can be characterized in terms of computability, primitive recursion, general recursion, lambda-definability, and inductive definition. Following Tomalin and Bar-Hillel (1953), this paper accepts the definition of recursion as based on the process of inductive analysis. 2. There are many examples of the operation of recursion and iteration in biological systems outside of communication. The codes specified in the DNA have an embedded and hierarchical structure. Mechanisms and structures are repeated throughout our physiology. So, the issue at stake here is not the importance of recursion to biology, but whether humans are the only animals that utilize recursion in communication.

The emergence of linguistic complexity 

recursive insertion could go on forever, with sentences like (4) being expanded to (7) and so on.

(7) The boy who stole the bike that the teacher had painted has a black jacket.

In this example, the subject “boy” is separated from its predicate “has a black jacket” by intervening material that could, in turn, be expanded indefinitely. Of course, if this expansion goes on too long, we will loose track of the identity of the subject and eventually the sentence will become more and more difficult to interpret, as in (8)

(8) The boy who stole the bike that the teacher Mary saw at the train station had painted has a black jacket.

Center-embedded structures become even more difficult when they are composed of common nouns with definite articles and confusable verbs, as in

(9) The dog the cat the boy the girl liked saw chased ran.

In practice, it is often difficult to distinguish recursion from iteration. Consider sentences such as (10) and (11). (10) Mary invited Tim’s mother to dinner. (11) Mary invited Tim’s mother’s friend to dinner. (12) Mary invited Tim’s mother’s friend’s family to dinner. The process of expanding these possessive phrases can be viewed as either recursion as in (13) or as iteration as in (14) (13) Mary invited (((Tim’s mother’s) friend’s) family) to dinner. (14) Mary invited Tim’s + mother’s + friend’s + family to dinner. In logical terms, we might view (13) as a superior representation. However, because interpretation is incremental, the resultant understanding may not be very different in the two cases. This contrast between iteration and repetition also arises for structures such as (15) with repeated complementation or (16) with right-branching relatives. (15) Mary said that Bill promised to tell Frank to ask Sarah to suggest to … (16) This is cat that chased the rat that scared the boy who …. There is an important difference between right-branching sentences like (13) through (16) and center-embedded cases like (7) through (9). The right-branching structures place little burden on working memory, since there are no discontinuous grammatical relations. Instead, words can be assigned to correct grammatical roles in an incremental fashion, as they occur. With center embedding, on the other hand, the embedded material directly interferes with incremental assignment of grammatical roles, placing

 Brian MacWhinney

a major burden on working memory. Similar problems occur with left-branching structures, such as those found in languages like Japanese that place the relative clause before the noun. In general, it seems that an iterative account for the generation of complex structures makes most sense when there is no interruption of constituents, suggesting that recursion involves more than just syntactic patterns. These facts show that neither recursion nor iteration could plausibly be based on any single simple neuronal device. In fact, we will see that the infinite complexity of human language arises from the interplay between at least six major linguistic subsystems.

1. Six subsystems — six theories We can begin this exploration by recognizing the importance in both linguistic and neural terms of the six basic levels of linguistic analysis: auditory phonology, articulatory phonology, lexicon, syntax, phrasal storage, and discourse. The account provided here attributes the computation of recursion to mechanisms operating within each of these six levels or subsystems. Table 1 presents an overview of this levels-based analysis of the emergence of recursion. Subsystem

Area

Processes

Theory

Audition

Auditory cortex

Extracting units

Statistical learning

Articulation

IFG, motor cortex

Targets, timing

Resonance, gating

Lexicon

Wernicke’s

Phonology to meaning

DevLex

Syntax

IFG

Slots, sequences

Item-based patterns

Storage

DLPFC

Attachment, roles

Mental Models

Dorsal cortex

Binding, competition, lists Deixis, perspective

Perspective

The core of the argument being developed here is as follows. The first three levels work together to construct a system for lexical learning and processing. To build this system, the child must work for years to extract units through audition (Jusczyk, 1997), to link articulation to audition (Oller, 2000), and to establish articulatory fluency. Without the consolidation of these peripheral processes, lexical processing would be impossible. Within the lexicon, self-organization links the topographic structure of posterior areas (including Wernicke’s area) to processors in the inferior frontal gyrus (IFG) (Grodzinsky & Friederici, 2006). The architecture of these processors can produce generativity through iteration and recursion, but without the further systems of storage and mental models, their operation would be vacuous. Storage mechanisms keep words and phrases in short term memory while the Merge operation continues to mediate between syntax and the lexicon. This processing involves a three-way

The emergence of linguistic complexity 

coordination. This coordination is then further modulated by the construction of embodied mental models (MacWhinney, 2008b) that can then offload phrases from temporary storage. This analysis shows that linguistic complexity, including recursion, requires the complete participation of all six of these linguistic levels. The development of this system was not a sudden, nearly accidental evolutionary event. Rather, the system has built upon six million years of development in a unique evolutionary niche occupied by a species that combined upright posture with tight, but competitive, social organization (MacWhinney, 2008a).

2. Systems vs. modules Before exploring in detail the processes and structures operative on each of these six subsystems, we need to consider some basic issues regarding pseudo-modular organization in the brain. It is tempting to think of these six subsystems as computational modules (Fodor, 1983; Levelt, 1989; Pinker, 1997). Indeed, each of these subsystems utilizes uniquely adapted processors in localized brain regions. However, thinking of these processors as protected modules like those in a Java program is not in accord with what we know about the brain (Bullinaria, 2007). In fact, brain regions are heavily interconnected by asymmetric bidirectional connections. These connections cannot pass symbols, as required by the digital computer. Instead neurons must communicate by depending on isotopic mapping, learned patterns of connectivity, firing synchrony, and modulation through supervisory units. Moreover, except in lower organisms or the brainstems of higher organisms, it is seldom the case that a single cell is responsible for a discrete cognitive function. This is certainly true at the level of the cortex, where cells appear to operate in assemblies of thousands of neurons to achieve single cognitive goals. Brain development in the fetus involves the migration of neurons from the germinal matrix to the periphery. During this migration, cortical areas maintain their connections to the various subcortical areas from which they differentiated. For example, within both the thalamus and the hippocampus, there are separate nuclei that project to separate cortical areas. Although single axons fire in a directional fashion, these larger sets of connections are bidirectional, thereby providing a system of reentrance or interaction between areas of the brain (Tucker et al., this volume). Within each of these thalamic or hippocampal nuclei there may be additional fine-grained structure that allows the subcortical area to maintain a map of the structure of the cortical area, even after it has migrated to a more distal position. The map-like nature of these connections between cortical and subcortical structures is further supplemented by map-like connections of motor and sensory areas to external sense organs and the body. In motor

 Brian MacWhinney

cortex, there is a somatotopic organization that matches up well with the actual shape of the human body. In sensory areas, cortex is organized to represent the features of the sense. For example, auditory cortex is organized in terms of frequencies, as detected by the neighboring cells in the cochlea. Similarly, the visual cortex is organized in patterns that maintain the position of receptors in the left or right visual field and other peripheral patterns. This map-like organization of the brain allows areas to communicate in terms of an embodied neural code that is implicit in the position of a neuron within the map. As processing moves away from the periphery, the blending of these codes increases. However, through reentrance, it is possible for the brain to ground cognition on these original body maps (Jeannerod, 1997; Schütz-Bosbach & Prinz, 2007; Wilson & Knoblich, 2006). This basic principle of map-like organization also applies to learning methods that function to organize local maps. One powerful way of modeling this local organization relies on the self-organizing feature maps (SOFM) of Kohonen (1990). In this model, neuron-like units are organized in two-dimensional sheets with connections to an array of input and output features. When an input feature vector is activated, units in the map also gain some activation. Through lateral inhibition, the most strongly activated unit will inhibit its neighbors, leading to a winner-take-all effect. This pattern of activity has been well documented for cortical structures. After the initial inhibition, there is then a learning phase in which the connections with the winner and its neighbors are strengthened. As a result of this learning, responses to certain patterns in the input tended to become parceled out across areas of the feature map, with this selforganized differentiation increasing over time. This type of map is a sparse, distributed memory, since there are typically many possible features of which only a few are active for a given input.

2.1 Subsystem 1: auditory organization Research on the emergence of auditory processing abilities has identified three major streams of auditory learning and development. The first stream involves the core perceptual features of hearing. These features are derived from the anatomy and physiology of the ear, the cochlea, the auditory nerve, and the auditory cortex. These peripheral structures are shared in common by humans and other mammals. For example, Kuhl and Miller (1975, 1978) demonstrated that chinchillas are able to process the contrast between voiced stops like /b/ and unvoiced stops like /p/ in a categorical, humanlike way. The second stream of auditory development is also shared, at least in part, with our primate cousins. This is the ability to process the statistical regularities of spoken language. There is now abundant evidence that both infants (Aslin, Saffran, & Newport, 1999; Marcus, 2000) and monkeys (Hauser, Newport, & Aslin, 2001) are able to use sequential statistics to extract background properties of their

The emergence of linguistic complexity 

language involving prosodies, phonotactics, and possible segments. This learning allows infants to orient to speech and to develop codes for storing sequences of syllables. The third stream of auditory development involves the actual storage of sounds as potential candidate words. By six months, children are able to respond to their own name. Using the auditory codes they have developed, they can retain large numbers of traces of words they have heard, although often they have not yet learned what these words might mean.

2.2 Subsystem 2: articulatory organization Each of the three streams of auditory processing discussed above could, in theory, proceed without reference to the articulatory system. However, the human phonological system emerges through a process of vocal imitation that couples audition to articulation. Before infants can display real imitation of conventional forms, they must first practice and solidify their control over the basic production processes of speech production. Oller (2000) shows that, during the first six months, the infant is busy consolidating control over the use of the glottis to achieve consistent phonation. During this period, the infant works to control volume, pitch, and the overall contour of vocalizations. Once this achievement is consolidated, the next goal is the linkage of phonation to articulation by control of the jaw and the oral articulators. This linkage develops during the period of repetitive babbling, as the infant works to develop methods for consistently producing target syllables and syllable sequences. As Macneilage & Davis (2000) have noted, the predominant template in babbling is the CV syllable that has a structure that derives from the close-open lip-smacking gesture in other primates. Using this basic gesture, infants begin to forge a resonant linkage or coupling between the processes of articulation and audition. Piaget (1954) described this linkage as a circular reaction. Basically, the child is learning how to control articulation in order to reliably produce important target sounds. During the last months of the first year, babbling continues to deepen the linkage between articulation and audition. While this is happening, the child is paying ongoing attention to the encoding of statistical regularities in the input language. The linkage of articulation to audition then places a further set of constraints onto the ways in which patterns are detected, since the infant is now trying to actually imitate aspects of the target language. For example, by the end of the first year, the child learning a tone language such as Mandarin will be trying to develop ways of producing each of the four Mandarin tones. There are now several well-articulated models of the development of linkages between articulation and audition. These models (Guenther, 2006; Lacerda, 1998; Westermann & Miranda, 2004) emphasize the ways in which units in auditory space come to be constrained by the possible articulations produced by the child.

 Brian MacWhinney

Lindblom et al. (in press) argue that, to work effectively across contexts, this learning has to rely on the fact that the brain is able to compute motor equivalence between targets. This allows the learner to link together a targeted sound with an auditory effect, even when the actual details of the sound production vary radically across contexts. During this learning, the ambient language is providing ongoing input to auditory organization and this input provides new targets for the process of articulation. In the end, the structures encoded in auditory space become the strongest forces in this coupled system.

2.3 Subsystem 3: lexical organization Li, Zhao, & MacWhinney (2007) have developed a model of lexical learning based on a self-organizing features model (SOFM). This model, called DevLex, uses three separate self-organizing feature maps for auditory phonology, articulatory phonology, and lexical structure. In effect, DevLex provides us with a fully implemented, neurologically grounded, empirically supported account of organization for these first three linguistic subsystems. Featural organization on the DevLex auditory map relies on the PatPho representational system that parcels out segments into an autosegmental grid. For example, the word stoops is composed of an /st/ onset, the vowel nucleus /u/ and the coda /ps/. Within each of these components there are three slots for phonemes and each phoneme is coded in terms of a set of distinctive features. This model assumes that initial auditory processing has yielded a set of perceptual features that are associated with specific syllables, and slots (onset, nucleus, coda) within syllables. The PatPho system was introduced in MacWhinney & Leinbach (1991) and most subsequent work in neural network modeling of input phonology has used this framework. The activation of segments or syllables in a self-organizing feature map is further controlled through a sequence detection mechanism that expresses the form of a word as a linear trajectory through points in the feature map. Multiple positional variants of a given segment are represented as multiple neighboring nodes in the input phonology map. Because this map relies on the same SOFM framework used by the DIVA (Guenther, 2006) model, the DevLex account of phonological learning is generally compatible with DIVA. Output phonology is also represented through sequence control units that activate articulatory gestures organized in a second motor feature map. Figure 1 below shows the shape of DevLex. The three separate maps of the DevLex model represent three of the six core linguistic modules. These modules are each located in separate brain regions, connected by axonal projections. DexLex trains these connections using Hebbian learning. However, we will see later that there is reason to believe that other processes are involved. Input phonology is processed in the auditory cortex of the superior temporal sulcus. Output phonology is controlled by parts of Broca’s area, along nearby regions

The emergence of linguistic complexity  Word Meaning Representation (WCD) Self-Organization Semantic map (SOM) Hebbian Learning (Comprehension) Input phonology map (SOM) Self-Organization Word Form (Phonetic features)

Hebbian Learning (Production) Output sequence map (SARDNET) Self-Organization Word Sequence (Phonetic features)

Figure 1. The architecture of the DevLex system.

with motor cortex. The core semantic or lexical map is centered in Wernicke’s area, although it is actually far more generally distributed, as we will see later. Looking first at the control of input phonology, we know that this processing is focused in primary auditory cortex. This area, which spans Brodmann areas 41 and 42, lies in the posterior half of the superior temporal gyrus and the transverse temporal gyri or Heschl’s gyri. Within this area, there are in fact multiple tonotopic maps, each of which appears to represent a different view or processing slant on the whole range of the frequency spectrum. Work with rhesus monkeys has shown that the auditory system involves three levels of auditory processing with 15 different tonotopic maps. This pattern of multiple parallel isotopically organized maps is similar to the pattern of multiple parallel maps found in the motor system. Like many other cortical areas, the auditory cortex is also connected to its own specific thalamic nucleus, the medial geniculate nucleus, from which it receives input. Human auditory processing is fundamentally similar to that of other mammals and even birds. This suggests that the peripheral auditory system and the basic neural engine for auditory feature detection were consolidated prior to the evolution of hominids. Further neural processing can reduce the enormous complexity of the auditory world to a much smaller set of contrasts that can link to output phonology and lexical structure. This reduction of complexity is operative in other mammals. However, it is likely that, under the influence of linkage first to articulation and then to a lexicon,

 Brian MacWhinney

these categorizational processes extend further and occupy additional neural machinery in humans. In addition to an overall sharpening of contrasts, human and primate audition must also differ in the extent to which they rely on mechanisms for sequence detection. Although syllables can be perceived as wholes, multisyllabic words need to be encoded in ways that associate sounds with syllable position. Prosodic features, such as syllabic stress or moraic timing, can facilitate and sharpen this encoding, but it is likely that some form of sequence detection is involved in the interfacing of auditory processing with lexical recognition. These sequence detection processes may be present in other mammals, but they are probably elaborated in humans.

3. A distributed lexicon Lexical representations are far more distributed that the representations of input and output phonology. The broad area of cortex at the intersection of the parietal, occipital, and temporal lobes has further access to wide areas of the whole cortex. Unlike the feature maps for input and output phonology, the core conceptual lexicon must make contact with a very diverse set of connections across the brain. For example, words for tools must make contact with the motor gestures and postures involved in the use of these tools. Words for fruits must make contact with the visual properties of these flowers, including colors, shapes, and smells. Words for actions must make contact with the motor sequences, perceptual changes, and object affordances involved in these actions. The choice between closely related meanings are determined by competitions within the relevant areas. For example, motor and parietal areas involved in tool usage can determine the competition between alternative tools such as such as screwdriver vs. drill. The competing cell assemblies within the lower level of this hierarchy then transfer activation back to higher-level units in the central lexical map that connects to phonology. Figure 2 presents a sketch of how this hierarchical organization can operate within a system of self-organizing feature maps (Dittenbach, Rauber, & Merkl, 2002). To control this hierarchical access, the brain must rely on long-distance connections between the core lexical areas and areas that flesh out the meanings involved in words. Moreover, these hierarchical connections must be structured in a way that allows for a consistent control of competition at both the local areas and the lexical core.

4. Consolidation of lexical patterns This distributed, hierarchical patterning has important consequences for the consolidation of linguistic complexity. Tucker et al. (this volume) argue that ventral stream

The emergence of linguistic complexity  layer 0 layer 1

layer 2

layer 3

Figure 2. Hierarchical access in self-organizing feature maps.

processing operates upon discrete item-based object representations that are characteristic of processing in temporal cortex. This type of item-based encoding is supported by neostriatal attentional mechanisms and hippocampal reentrant encoding processes. The hippocampus provides a compressed encoding of the distributed patterns related to a word. By maintaining resonant and reentrant reactivation of these patterns, the hippocampus can facilitate the consolidation of these traces into a new cell assembly or lexical pattern. This ventral-hippocampal system provides the basic engine for consolidating and extending linguistic complexity at the lexical level. Here the complexity involves not just the phonological form of the word, but also the diverse connections of the lexical system to many areas of the brain. Because words have become conventionalized mimetic forms, this system then functions to repeatedly consolidate variant meaningful configurations into the same phonological bucket. From this core engine, arise the linguistic complexities of radial semantic structure (Lakoff, 1987), polysemic pathways (MacWhinney, 1989), metonymy, partonomy, and homonymy (Lyons, 1977). Underneath this linguistic complexity, there is a further level of psycholinguistic complexity that arises from the distributed, hierarchical shape of lexical activation. A simple word like “hammer” is able to trigger both visual images of a hammer in the ventral “what” stream and functional images of wielding a hammer to hit a nail in the dorsal “how” stream. When we come to more complex words such as “grandfather” or “promise”, the meanings involved have to be unpacked in terms of a whole set of

 Brian MacWhinney

embedded predicates, such as “the father of my father, or the father of my mother” or “tell someone that you will perform an action that you would not otherwise have done with the expectation that, if you fail to complete the action, there would be unpleasant social or interpersonal consequences and that you therefore fully intend to complete the action, even if certain barriers arise.” The solidification of complexity at the lexical level relies on this system of distributed resonance. When we produce a complex word like “promise”, it is not necessary that all elements of the chain be fully activated in working memory. All that is necessary is that enough of the word be activated to guarantee correct lexicalization of this word as opposed to its competitors. In the case of “promise” it may be that all that is necessary is the notion of saying something seriously. The further pragmatic implications of the word may not be available initially in working memory. However, there are long-term, distributed links available that can call them up if needed. This is equally true for concrete terms such as “hammer”, since we do not always have to think about using the claw of a hammer to pull out a nail when we hear the word “hammer.” In this way, we can think of words as promissory notes or tokens that are issued in the place of the full set of concepts and stances with which they are linked.

5. Sequence analysis within the lexicon In principle, it would be possible to ground a communication system on sentences or propositions compressed into single words. Polysynthetic languages such as Iroquois or Eskimo push hard in this direction with their inclusion of a wide range of moods, persons, surfaces, and aspects into a single verb-based complex. However, languages achieve this compression by relying on a variant method for controlling the sequencing of lexical items. Consider the contrast between Hungarian and English in the way they form the phrase meaning “my coat.” In English, the possessive appears as a separate word preceding the noun. Variations in the phonological shape of the following noun have minimal effect on the sound of the possessive pronoun. In Hungarian, on the other hand, the suffix -om takes on the shape of -am, -om, -em, -öm, or -m, depending on the shape of the stem. Moreover, the stem changes its shape, depending on the nature of the suffix. The debate about the cognitive representation of these morphophonological patterns has raged for over three decades in psycholinguistics. The connectionists and analogists view forms such as kabátom as produced within the lexicon through interactive activation of analogic patterns. In this model, all lexical forms are produced within the lexicon, without reliance on external routes. The alternative view holds that regular morphological forms are produced by combination between stems and affixes. The third possible formulation is that of MacWhinney (1978, 1982, 1987a, 2005) which

The emergence of linguistic complexity 

views combinatorial forms as arising through extraction from a core analogic process. Within the framework of self-organizing feature maps, this means that a separate lexical map for affixes emerges from a process that compares similar morphological formations. For example the comparison of shoe with shoes will lead to the extract of -s as the initial productive form for the plural. Similarly, the comparison of kabát with kabátom leads to the extraction of -om as the first person possessive suffix. This comparison and extraction method is clearly an important additional source of linguistic complexity. This same engine can work within the noun phrase or verb phrase to extract my from the combination my coat, just as -om is extracted from kabátom. Moreover, it is an engine that can work in both directions. If the pressures of fast speech work to modify combinations such as going to into gonna, then the latter can be stored as a single form representing what was earlier a syntactic combination. The extreme analogist view would hold that the neurological basis of morphology is completely interwoven with the lexical substrate in Wernicke’s area at the juncture of the parietal and temporal lobes. It would view an item such as -om as occupying the same lexical map as an item such as kabát. One strength of this approach is that the morphological alterations involved in the complex word can be directly controlled through connections within the DevLex maps. However, a weakness in this approach is that it fails to capture the fact that the -om suffix occurs positionally after the stem. This problem becomes particularly severe in agglutinative languages where strings of two, three, or even more suffixes appear after the noun. To control these processes within a single local net, sequence detector units would have to be built into the lexical net itself. As an association area, Wernicke’s is not specialized for this type of sequential processing.

5.1 Subsystem 4: positional patterns To solve this problem and to boost lexical capacity, evolution turned over combinatorial processing to Broca’s area in the inferior frontal gyrus (IFG). Among the various cortical areas specialized for sequence processing, this is the area that was closest to the posterior lexical areas. Grodzinsky & Friederici (2006) argue that the processing of positional patterns depends on connections between the anterior superior temporal gyrus (STG) and the inferior portion of Broca’s area lying in the pars opercularis (BA44). They distinguish this circuit from a parallel, but more dorsal one between posterior STG and the superior portion of Broca’s area in the pars triangularis (B45). Figure 3 below displays these two circuits. The basic connectivity between these two circuits is already present in monkeys (Deacon, 1989). Therefore, there was no initial need to establish connectivity between the areas. In this sense, the syntactic engine was not built up from scratch. Rather, like all evolutionary advances, these circuits comprise a new machine made

 Brian MacWhinney

Structural connectivity: Tractography data (DTI)

Subject 1

Subject 3

Subject 1

Subject 3

Subject 2

Subject 4

Subject 2

Subject 4

from FOP to STG via the fasciculus uncinatus

from BA44/45 to STG via the fasciculus longitudinalis superior

Figure 3. Anterior-posterior tracts measured through DTI. Source: Friederici, Bahlmann et al., PNAS, 2006.

up of old parts. Within this new machine, there was a need to make sure that this connectivity supported effective control of lexical activation. To do this, it was important for lexical assemblies in posterior cortex to organize themselves in ways that map up with the already existent connections to IFG. Again, this is not some sudden evolutionary invention, but rather the reshaping of an old machine to serve new functions. The linkage of posterior lexical areas to IFG relies crucially on the establishment of isotopic maps between the two areas. The DevLex model shows how the topological structuring of posterior cortex is achieved through movement of lexical forms on the self-organizing feature map. Figure 4 below illustrates the results of training the DevLex model on parental input derived from the Belfast corpus in CHILDES (MacWhinney, 2000). During this training, words that appear in similar contexts in the parental input self-organize so that they end up being located next to each other in lexical space. In other words, nouns end up next to other nouns and prepositions end up next to other prepositions. This topological self-organization provides support for reliable interactions between syntactic patterns in IFG and the posterior lexical map. In effect, the topological organization of the map is the backbone of a communication protocol between the lexicon and IFG. To understand how this protocol operates to produce complex syntactic structures, we will need to take an excursion into language acquisition theory.

The emergence of linguistic complexity 

Figure 4. Self-organization in the DevLex model.

6. Item-based patterns In the early days of acquisitional theory, Braine (1963, 1971) explored ways of applying learning theory to the study of child language. The formulation he devised focused on the idea that function words tend to appear in fixed positions vis a vis content words. For example, the appears before nouns and the suffix -ing appears after verbs. Like Harris (1951), Braine analyzed these constituent structures in terms of slots that could be filled by items of a certain class. Formulating a set of 12 such rules for a small corpus of child utterances, he referred to his account as a “pivot-open” grammar, since it specified the position of pivot words vis a vis the open class. Under the influence of Chomsky’s (1957) ideas about deep structure, this model was criticized as failing to pay adequate attention to semantic patterning. Later, Braine (1976) revised his account, emphasizing the role of “groping patterns” that established links based not on lexical class, but semantic relations. Sticking more closely to Braine’s original formulations, MacWhinney (1975) introduced the notion of the item-based pattern. Applying this construct to a corpus of Hungarian, MacWhinney examined the word order of 11,077 utterances produced by two Hungarian children between the ages of 17 and 29 months. He found that between 85 and 100% of the utterances in these samples could be generated by a set of 42 item-based patterns. Some examples of these patterns in English translation are: X + too, no + X, where + X, dirty + X, and see + X. The item-based

 Brian MacWhinney

pattern model was able to achieve a remarkably close match to the child’s output, because it postulates an extremely concrete set of abilities that are directly evidenced in the child’s output. MacWhinney made no general claims about a pivot or open class, focusing instead on the idea that the first syntactic patterns involve links between individual lexical items and other words with which they are prone to combine. An example of an item-based pattern is the structure the + X. This pattern states simply that the word the occurs before another word with which it is semantically related. In addition to these positional facts, the item-based pattern encodes the shape of the words that can occupy the slot determined by X and the nature of the semantic relation between the and X. This is to say that an item-based pattern is a predicate-argument3 relation which encodes: •• the lexical identity of the predicate, •• the lexical category of the argument(s), •• the sequential position of the predicate vis a vis its argument(s), and •• the semantic relation between the predicate and its argument(s). During language learning, the child generalizes item-based patterns to produce feature-based patterns or constructions. Feature-based patterns are no longer linked to specific lexical items, but instead apply to classes of items. From item-based patterns like my + X the child can extract the feature-based pattern Possessor + Possession. In this way, the child slowly pieces together the 23 major grammatical dependency relations of English, as summarized in the work on the GRASP parser (Sagae, Davis, Lavie, MacWhinney, & Wintner, 2007) for the CHILDES database. In this system, predicates can attach to as many as three arguments. Item-based constructions for verbs can also include the verbs of embedded clauses as arguments. And we will see below how itembased constructions for prepositions and auxiliaries include both an endohead and an exohead.

3. This paper uses the predicate-argument relation to describe item-based dependency patterns. This terminology is used to avoid confusions regarding the ways in which clusters inherit head features for X-bar syntax. In the noun phrase, predicates join with their heads to produce new clusters that inherit the features of the head noun. However, in verb phrases and prepositional phrases featural inheritance is driven by the predicate, not the arguments. Because of this, referring to the arguments as the head of a verb phrase would be confusing. The major danger involved in use of predicate-argument terminology for item-based patterns is the possibility that this would be interpreted as applying outside the domain of lexical combinations. Other levels of predicate-argument decomposition and combination exist throughout language and cognition and we are here only focusing on the role of the predicate-argument relation for combinations of words.

The emergence of linguistic complexity 

There is a third level of argument generalization, above the levels of the itembased pattern and the feature-based pattern. This is the level of the global construction. Just as feature-based constructions emerge from a process of generalization across item-based patterns, so global constructions emerge from generalization across feature-based constructions. For example, in English, there are literally dozens of verb groups that share a common placement of the subject before the verb. Together, these constructions give support for the SV global construction in English. The SV and VO global patterns of English work together to produce prototypical SVO order (MacWhinney, Bates, & Kliegl, 1984). Other languages promote different combinations of global patterns. In Hungarian and Chinese, for example, SV, OV, and VO orders operate to express alternative varieties of object definiteness, producing SVO and SOV orders. Italian combines SV and VO patterns with secondary, but significant use of VS (Dell’Orletta, Lenci, Montemagni, & Pirrelli, 2005) to produce SVO and VSO orders. Other global patterns control the ordering of topic before comment or the tendency to associate animacy with agency. In addition to this process of generalization, positional patterns can be subjected to a process called composition. Composition takes two positional patterns and hooks them up into a single larger sequence. The important consequence of composition is that it increases the proceduralized nature of syntactic processing. For example, it may be that a single complex network, looking very much like a finite state automaton, processes all variants of noun phrases. In this network, there would be an initial slot for a quantifier, followed by a determiner or possessive, then a series of adjectives, and finally the noun. The compilation of smaller patterns into larger patterns of this type can proceduralize and facilitate both listening and production.

6.1 Subsystem 5: storage Dependency grammars, such as GRASP, use structures equivalent to the IFG pattern detectors of the type outlined here. However, by themselves, dependency relations are not enough to achieve parsing or generation of longer strings of words. Some additional control mechanism is needed to allow for iteration or recursion. Here, there are two neurologically plausible approaches. One mechanism would compose X-bar structure and trees directly within IFG. However, neurological evidence for such embedded groupings of sequence processors is currently absent. Instead, current evidence suggests that areas outside of IFG are involved in the construction of larger conceptual trees from the sequential fragments detected by IFG. In accounts such as MacWhinney (1987b) or Gibson (1998), smooth processing relies on the incremental construction of interpretable units. Consider a sentence, such as my coat has a missing button. As soon as the sequence my coat is detected, the predicate is linked to its argument and the whole is then treated as a single cluster in the mental model being constructed.

 Brian MacWhinney

Mental model construction proceeds in accord with the principle of starting points introduced by MacWhinney (1977) and supported in detail by Gernsbacher (1990). The starting point of my coat then becomes the perspective from which the rest of the sentence is interpreted. At this point, resonant activation involves the activation of items in posterior lexical space, activation of the meaning components of these items, and continued processing in IFG. Next, the sequential processor takes this whole active assembly as input to the verb-based frame for have. This predicate has argument slots for both a possessor perspective and an object possessed. Even before the second slot is filled, incremental processing activates a mental model expectation for a thing possessed. This mental model structure is structured in terms of the theory of egobased perspective taking (MacWhinney, 2008b) described below. Next, the phrase a missing button is processed by the two relevant sequence processors and the result then fills the second slot of the verb has, thereby completing the mental model of a coat that has a missing button. Of course, the model itself may generate additional associated ideas. Perhaps the button is removed in some overt way; perhaps it is seen on the floor; or perhaps there is a focus on the thread left on the coat after the button has fallen off. The filling of argument slots in feature-based patterns is driven by a series of cues that have been studied in detail in the context of the Competition Model of MacWhinney (1987a, 1987b). The model specifies a series of steps for the ways in which incremental processing triggers competition between constructions: •• Sounds are processed as they are heard in speech. •• Competition during sound processing controls activation of a current word. •• Each new word activates its own item-based patterns along with related featurebased patterns (see below). •• Item-based patterns then initiate tightly specified searches for slot fillers. •• Slots may be filled either by single words or by whole phrases. In the latter case, the attachment is made to the head of the phrase. •• To fill a slot, a word or phrase must receive support from cues for word order, prosody, affixes, or lexical class. •• If several words compete for a slot, the one with the most cue support wins. Most work on the Competition Model has focused on comprehension, which is easier to control experimentally. However, the model applies equally well as an account for sentence production. The details of the operation of this parser are controlled by the competitions between specific lexical items and the cues that support alternative assignments. Consider the case of prepositional phrase attachment. Prepositions such as on take two arguments; the endohead is the object of the preposition, the exohead is the head of the prepositional phrase (i.e., the word or phrase to which the prepositional phrase attaches). Consider the sentence the man positioned the coat on the rack. Here, the endohead of on is rack and its exohead could be either positioned or the coat.

The emergence of linguistic complexity 

These two alternative attachment sites for the prepositional phrase are in competition with each other. For detailed examples of the step-by-step operations of this type of processor consult MacWhinney (1987a), MacDonald, Seidenberg, & Perlmutter (1994), or O’Grady (2005). In this model, syntax involves nothing more than the repetitive clustering of the results of basic linear detectors. Of course, not all sentences are as simple as the one chosen to illustrate the basic process. Often uninterpreted arguments will build up on sentence memory waiting for merger with their predicates. MacWhinney & Pléh (1988) suggested that the capacity of memory for uninterpreted phrases was no greater than three and Gibson (1998) and others argue for a similar limit. But all analysts agree that there must be a mechanism for storing at least two or maybe three such uninterpreted items during processing. Because of its role in the phonological loop and other memory processes, there is reason to believe that dorsolateral prefrontal cortex (DLPFC) provides the necessary store for not-yet-merged items. This frontal mechanism then provides an additional engine for the maintenance and diversification of linguistic complexity. But can a mechanism like this really control complex syntax? Don’t we need the full power of transformational grammar, or at least context-sensitive phrase structure grammars? What about empty categories, traces, indices, interfaces, and so on? It is certainly important to explain how a linear mechanism of this type can compute complex structures. First, the basic generative power in this system comes from the fact that the slots of positional patterns refer to whole classes of items. Because of this, the power of the positional pattern machine is beyond that of finite-state processors that operate only on terminal symbols. As Hausser (1992) has shown, finite state grammars that operate on category symbols are formally equivalent to phrase-structure grammars. Second, the results of individual linear patterns can be combined or clustered through attachment in mental model space. As a result of this, the final model encodes full X-bar structure. Third, many of the linguistic phenomena that have been used to motivate complex syntax are actually better represented through memory processes in mental models, as discussed immediately below. Consider the case of the tangled dependencies caused by Dutch serial verbs or the English “respectively” construction. The fact that John and Bill ordered steak and fish, respectively can be interpreted best by a mnemonic device that establishes actual spatial positions in mental model space for John and Bill and then engages in the mental action of parceling out steak and fish to these positions in mental model space. This type of mental model processing is basic for anaphoric processing. There is no reason not to think that it is used to process these constructions too. Of course, the problem here is that, by itself, the syntax would not yield a complete parse tree in such cases. But that is because syntax is not doing this work alone.

 Brian MacWhinney

6.2 Subsystem 6: mental models Recent work in neuroscience has benefitted from four fundamental insights, each relating to the construction of mental models. First, in the 1980s, we learned that the visual system separates processing into an image-oriented ventral stream and an action-oriented dorsal stream.4 Second, we have learned from imaging work through the last decade that the brain relies on a perception-action cycle to interpret incoming messages. This cycle involves the generation of mental representations for objects in terms of the ways in which we typically act upon them (Knoblich, 2008). Much of this cycle is grounded on interactions that include the action-oriented processing of the dorsal stream. Third, we have learned that the brain provides specific mechanisms for mapping the body images of others onto ours. One consequence of this ability is the fact that “mirror” neurons (Rizzolatti, Fadiga, Gallese, & Fogassi, 1996) controlling actions, facial gestures, and postures can fire equally strongly when the actor is the self or the other. As we are now learning, these mirror systems are just one part of a general system for social cognition. The larger system also includes mechanisms for temporal facial processing (Pelphrey, Morris, & McCarthy, 2005) and eye contact (Pelphrey et al., 2003), as well as amygdala and striatal areas for empathy (Meltzoff & Decety, 2003) and projection (Adolphs & Spezio, 2006). Fourth, we have learned that the basal ganglia and hippocampus play a central role in the consolidation of memories, often driven by rewards and error minimization. Piecing together these results, and following Tucker et al. (this volume), we can see that one of the additional consequences of the dorsal-ventral dichotomy is a shift of discrete processing of individual elements to the ventral stream and a shift of global model construction to the dorsal stream, with particular additional regulatory control from frontal areas. In recent papers (MacWhinney, 2005b, 2008b), I have suggested that this system provides the neurological basis for a system that constructs dynamic mental models from linguistic input. At the core of this system is the notion of the self as actor. During sentence interpretation, this fictive self is then projected onto the role of sentence subject, and the self reenacts the image underlying the sentence. These

4. Following Givon (1995), Hurford (2002) relates the separation of processing into the dorsal and ventral streams to the predicate-argument distinction in language. However, as Bickerton (2002) notes in his commentary to the Hurford’s article, this analysis fails in two important regards. First, predicates and arguments are not “raw sensory feeds” but rather complex lexical items that can themselves involve embedded predications, as we noted earlier in our discussion of words like “promise” or “grandfather”. Second Hurford’s model fails to provide a method by which the brain can integrate predicates and arguments. The mechanism proposed in the current paper is not linked in any clear way to the dorsal-ventral contrast, depending instead on interactions across IFG, distributed lexical processing, and frontal mechanisms for mental model construction.

The emergence of linguistic complexity 

images place the self into a set of well-understood roles as agent, experiencer, and source. Even locative and temporal relations can be interpreted from the egocentric frame that begins with a projection of the self onto the objected located or the event in time. Because narrative and dialog often involve rapid shifts between agents, this system has to be able to use linguistic devices to control perspective shifting. As a result of this core dynamics, we can refer to this system as the Perspective Shift System. This system constitutes the highest level of support for linguistic complexity. Without the mental model construction supported by this system, complex syntax would be useless. This is because the fundamental purpose of virtually all the devices of complex syntax is the marking of perspective shift. This analysis applies across all the major grammatical constructions, including passivization, relativization, clefting, pronominalization, dislocation, existentials, shift reference, split ergativity, serialization, complementation, conjunction, ellipsis, adverbialization, long-distance anaphora, reflexivization, PP-attachment, and participial ambiguity. Each of these structures allows the speaker to combine, maintain, and shift perspectives in communicatively important ways. And these devices allow the listener to trace these movements of the speaker’s attention across all of these shifts.

7. Building mental models The traditional view of mental model construction (Budiu & Anderson, 2004; Kintsch, 1998) focuses on the linking of predicates into a coherent propositional graph This activity is much like the process of clause-combining that we learned in classes in composition. For example, you can combine “the dog chased the bird” and “the bird flew away” to form “the dog chased the bird that flew away.” All one needs here is a grammatical device that serves to mark the fact that the bird plays a role in both clauses. Language provides a variety of methods for clause linkage, including conjunction, complementation, relativization, subordination, and adverbialization, as illustrated in (17)–(22): (17) You go down to Shattuck, and the bakery is on the corner. (18) Jim wanted me to bring him a loaf of bread. (19) The bread I bought was unfortunately stale. (20) Although it was stale, John wanted to pay me. (21) Shaking my head, I accepted the money. (22) My refusal of the payment would have made him upset.

 Brian MacWhinney

These different methods link together clauses in terms of spatial, temporal, causal, and anaphoric relations. Most of these methods rely on both lexical and syntactic processes. Typically, there is a linking lexical element, such as a conjunction or relativizer. In some cases, the lexical element is an affix that forms nominalizations or participials. Use of this linking element then triggers additional syntactic processes, such as extraposition, deletion, constituent reordering, agreement, and so on. These processes rely on four key aspects of the current model. First, the lexical items involved in these constructions must be stored as phonological forms within the posterior systems. These items include both full lexical items and affixes. Second, these lexical forms must be integrated by IFG into positional patterns that control the positioning of the items in the clauses, as well as movement and deletion. Third, these IFG structures must rely on frontal STM mechanisms that store elements. Fourth, these STM items must be pieced together for final mental model construction. MacWhinney (2008b) argues that mental model construction is driven by a process of perspective taking. Let use consider an example from relative clause processing. Earlier we noted how clause combining through STM joins “the dog chased the bird” and “the bird flew away” to form “the dog chased the bird that flew away.” In this case, the shift moves smoothly from bird as the object of chased to bird as the subject of flew away. However, if the sentence is “the dog chased the bird that the girl loved” then the perspective shift is far more difficult, since a brand new perspective is introduced and the perspectives of both the dog and the bird must be dropped. In this case, there is a greater burden on STM for fragment storage and hence a higher overall processing load, as reflected by slower latencies and lesser recall accuracy for object relatives. These shifts of perspective are triggered by syntactic patterns linked to lexical devices. To learn these, the child must figure out how to operate on signals from the lexicon or IFG to control the correct shifting in frontal cortex. As the developmental literature amply demonstrates, the learning of this control takes many years (Franks & Connell, 1996). Later in this paper, we will explore some of these processes in further detail, since this is one of the primary loci of the consolidation of linguistic complexity. Ambiguous sentences illustrate another face of perspective shifting. Consider sentences such as “John saw the Grand Canyon flying to New York.” Here, the default syntactic mechanism would favor the local attachment of flying to New York to Grand Canyon. The competing attachment is to John. Of course, the latter perspective is far more plausible. Or consider the processing of “Visiting relatives can be a nuisance.” Here, we can either take the perspective of the relatives who become a nuisance to their hosts or the perspective of an unmentioned generalized actor who visits the relatives. In this case, both readings seem plausible. Reflexivity provides another useful example of perspectival processes. Consider these sentences

The emergence of linguistic complexity 

(22) Jessie stole a photo of herself/her* out of the archives. (23) Jessie stole me a photo of herself/her out of the archives. (24) Jessie stole a silly photo of herself/her out of the archives. In (a) the reflexive is required, because the perspective of Jessie remains active up to the appearance of the anaphor. In (b) on the other hand, the intervening presence of “me” causes a shift of perspective away from Jessie. As a result, when interpretation reaches the anaphor, either the reflexive or the simple pronoun is acceptable. Perspective shift is sensitive not just to other intervening animate perspectives, but also to implicit perspectives triggered by adjectives such as silly in (c). This type of phenomena is basic to all levels of mental model construction.

8. Perspective and Gesture The frontal-parietal system for perspective shifting is not a recent evolutionary adaptation. Chimpanzees (Tomasello, Call, & Gluckman, 1997), dogs, and other mammals make extensive use of symbolic behaviors in social contexts. However, lacking a lexicon and positional patterns, other animals cannot organize these behaviors into recursive structures.5 However, Donald (1991) and others have argued that the production of symbolic communication can rely on gestural and vocal devices that may well have been readily accessible to homo erectus. Because gestures can be formed in ways that map iconically to their referents, it is relatively easy to build up communal recognition of a gestural system. As Tucker et al. (this volume) argue, such a system would rely primarily on gestures and affordances specific to the action-oriented processes in the dorsal stream. It appears that speakers of sign languages are able to use posterior lexical areas to structure a lexicon of signs, just as they use IFG in the left hemisphere to control the ordering of signs. It is possible that protosign could also have relied on these same neuronal structures for lexical organization. However, looking back two million years, it is likely that the depth of support for lexical storage and positional patterning of gesture was still very incomplete. As a result, it is likely that protosign was incompletely lexical and heavily reliant on dorsal processes for direct perspective taking and shifting.

5. Gentner, Fenn, Margolish, & Nusbaum (2006) claim that starlings demonstrate recursive processing for strings such as AAABBB. However, Corballis (2007) points out that these strings can be detected through a subitization-based counting mechanism that has been demonstrated for birds, along with the ability to match two counts.

 Brian MacWhinney

Although sign may not have triggered full linguistic structure, it provided a fertile social bed that supported the development of further articulatory, lexical, and sequence systems. As Darwin (1872) notes, vocal and gestural communication coexisted as parallel streams from the beginning of human evolution. Gesture and prosody were able to keep humans engaged in protoconversations, during which the further elaboration of vocal patterns could refine and complement communication in the gestural-prosodic mode. Of course, humans are not the only primates that engage in conversation. However, as argued in MacWhinney (2008a), the shift in homo habilis to a full upright posture led to two important consequences. One was the freeing of the hands for additional conversational interaction and the other was the encouragement of full face-to-face interactions linked to full display of the hands and torso. This increasing support for gestural communication brought along with it a supportive social context for the further development of accompanying vocalizations. However, both of these modalities continue to provide important input to conversation in modern humans. Thus, we can best view the transition from a primarily gestural communication to a primarily vocal communication system as gradual, but unbroken, process with no sudden break based on the sudden introduction of an ability to process recursion.

9. Integration of the subsystems We have now finished our survey of an account of neurolinguistic processing grounded on self-organizing feature maps, sequence processing mechanisms, limbic consolidation, and topological preservation of feature map resonance across six linguistic modules. The core mechanisms of neural connectivity and firing are fundamental to all animals from molluscs to mammals. Mechanisms for sequence detection and control can also be found in both invertebrates such as insects and vertebrates such as amphibians. Systems of topographic organization can be found even in animals with no cortex. Systems controlling memory consolidation and value-based projection are found in bees (Menzel & Giurfa, 2001). What is new in the engines supporting language are not the pieces, but the ways in which the pieces are being combined. We are now in a position to contrast the sketch of the neurological bases of language processing provided here with the proposal advanced by HC&F. Their claim is that the Merge operation produces recursion and, hence, complex syntax. The problem with this proposal is that it identifies only one of the six subsystems required to support complexity and recursion itself. As we have seen, at least six subsystems must work together to produce complex syntax. Recursion and iteration are not produced within a single subsystem, but through the interaction of subsystems. Of course HC&F could argue that the other five systems are parts of FLB and therefore present in other species. But there is no evidence in other species for a linkage between audition and

The emergence of linguistic complexity 

articulation of the type seen in human babbling. Nor is the evidence for the presence of a lexicon capable of expressing tens of thousands of concepts, along with detailed shades in their meanings. Moreover, there is no evidence in other species of anything similar to the circuit between IFG and STM areas in DLPFC. Of the various systems supporting recursion in human language, the one that seems to be least changed by language evolution is the system for mental model construction. However, even here, we have evidence that humans far surpass their primate cousins in tests of perspectivetaking and theory of mind (Tomasello, Call, & Gluckman, 1997). The issue raised by HC&F is a good one, since even an incomplete answer to this puzzle forces us to focus on the core issue of how the brain generates linguistic complexity. Beyond recursion and iteration, complexity arises from processes of anaphora, deixis, derivation, affixation, compounding, prosodic packaging, phonological assimilation, and presupposition. The other chapters in this book discuss in great detail the contributions of these processes to language change. Like recursion and iteration, each of these processes relies on all six neurolinguistic subsystems for its full instantiation. Thus, it seems best to conclude that complexity arises from six permeable subsystems each designed to interact smoothly with the others. Each of the pieces of this integrated engine made important separate, contributions to human evolution. However, the full power of the system could only be realized once all six components were fully integrated.

References Adolphs, R. & Spezio, M. 2006. Role of the amygdala in processing visual social stimuli. Progress in Brain Research 156: 363–378. Aslin, R.N., Saffran, J.R. & Newport, E.L. 1999. Statistical learning in linguistic and nonlinguistic domains. In The Emergence of Language, B. MacWhinney (Ed.), 359–380. Mahwah NJ: Lawrence Erlbaum Associates. Bar-Hillel, Y. 1953. On recursive definitions in empirical science. In 11th International Congress of Philosophy, Vol. 5, 160–165. Bickerton, D. 2002. Afferent isn’t efferent, and language isn’t logic, either. Behavioral and Brain Sciences 26: 286. Braine, M.D.S. 1963. The ontogeny of English structure: The first phase. Language 39: 1–13. Braine, M.D.S. 1971. On two types of models of the internalization of grammars. In The Ontogenesis of Grammar: A Theoretical Symposium, D.I. Slobin (Ed.), New York NY: Academic Press. Braine, M.D.S. 1976. Children’s first word combinations. Monographs of the Society for Research in Child Development, 41, Whole No. 1. Budiu, R. & Anderson, J. 2004. Interpretation-based processing: A unified theory of semantic sentence comprehension. Cognitive Science 28: 1–44. Bullinaria, J.A. 2007. Understanding the emergence of modularity in neural systems. Cognitive Science 31: 673–695.

 Brian MacWhinney Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: The MIT Press. Corballis, M. 2007. Recursion, language, and starlings. Cognitive Science 31:697–704. Darwin, C. 1872. The Expression of the Emotions in Man and Animals. London: John Murray. Deacon, T. 1989. The neural circuitry underlying primate calls and human language. Human Evolution 4: 367–401. Dell’Orletta, F., Lenci, A., Montemagni, S. & Pirrelli, V. 2005. Climbing the path to grammar: A maximum entropy model of subject/object learning. Association for Computational Linguistics: PsyComp Models. Dittenbach, M., Rauber, A., & Merkl, D. 2002. Uncovering the hierarchical structure in data using the growing hierarchical self-organizing map. Neurocomputing 48: 199–216. Donald, M. 1991. Origins of the Modern Mind. Cambridge MA: Harvard University Press. Friederici, A.D., Bahlmann, J., Heim, S., Schubotz, R. I. & Anwander, A. 2006. The brain differentiates human and non-human grammars: Functional localization and structural connectivity. PNAS 103(7): 2458–2463. Fodor, J. 1983. The Modularity of Mind: An Essay on Faculty Psychology. Cambridge MA: The MIT Press. Franks, S.L. & Connell, P.J. 1996. Knowledge of binding in normal and SLI children. Journal of Child Language 23: 431–464. Gentner, T.Q., Fenn, K.M., Margoliash, D. & Nusbaum, H.C. 2006. Recursive syntactic pattern learning by songbirds. Nature 440: 1204–1207. Gernsbacher, M.A. 1990. Language Comprehension as Structure Building. Hillsdale NJ: Lawrence Erlbaum. Gibson, E. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition 68: 1–76. Givón, T. 1995. Functionalism and Grammar. Amsterdam: John Benjamins. Grodzinsky, Y. & Friederici, A. 2006. Neuroimaging of syntax and syntactic processing. Current Opinion in Neurobiology 16: 240–246. Guenther, F. 2006. Cortical interactions underlying the production of speech sounds. Journal of Communication Disorders 39: 350–365. Harris, Z.S. 1951. Structural Linguistics. Chicago IL: The University of Chicago Press. Hauser, M., Chomsky, N. & Fitch, T. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569–1579. Hauser, M., Newport, E. & Aslin, R. 2001. Segmentation of the speech stream in a non-human primate: Statistical learning in cotton-top tamarins. Cognition 78: B53–B64. Hausser, R. 1992. Complexity in left-associative grammar. Theoretical Computer Science 106. Hurford, J. 2002. The neural basis of predicate-argument structure. Behavioral and Brain Sciences 26: 261–316. Jeannerod, M. 1997. The Cognitive Neuroscience of Action. Cambridge MA: Blackwell. Jusczyk, P.W. 1997. The Discovery of Spoken Language. Cambridge MA: The MIT Press. Kintsch, W. 1998. Comprehension: A paradimg for Cognition. Cambridge: CUP. Knoblich, G. 2008. Bodily and motor contributions to action perception. In Embodied Cognition, R. Klatzky, B. MacWhinney & M. Behrmann (Eds), Mahwah NJ: Lawrence Erlbaum. Kohonen, T. 1990. The self-organizing map. In Proceedings of the IEEE, Vol. 78, 1464–1480. Kuhl, P.K. & Miller, J.D. 1975. Speech perception by the chinchilla: Voiced-voiceless distinction in alveolar plosive consonsants. Science 190: 69–72. Kuhl, P.K. & Miller, J.D. 1978. Speech perception by the chinchilla: Identification functions for synthetic VOT stimuli. Journal of the Acoustical Society of America 63: 905–917.

The emergence of linguistic complexity 

Lacerda, F. 1998. An exemplar-based account of emergent phonetic categories. Journal of the Acoustical Society of America 103: 2980–2981. Lakoff, G. 1987. Women, Fire, and Dangerous Things. Chicago IL: Chicago University Press. Levelt, W.J.M. 1989. Speaking: From Intention to Articulation. Cambridge MA: The MIT Press. Li, P., Zhao, X. & MacWhinney, B. 2007. Dynamic self-organization and early lexical development in children. Cognitive Science 31: 581–612. Lindblom, B., Diehl, R., Park, S.-H., & Salvi, G. In press. Sound systems are shaped by their users: The recombination of phonetic substance. In Where do features come from? The nature and sources of pho-nological primitives. Lyons, J. 1977. Semantics. Cambridge: CUP. MacDonald, M.C., Pearlmutter, N.J. & Seidenberg, M.S. 1994. Lexical nature of syntactic ambiguity resolution. Psychological Review 101(4): 676–703. MacNeilage, P., Davis, B.L., Kinney, A. & Matyear, C. 2000. The motor core of speech: A comparison of serial organization patterns in infants and languages. Child Development 71: 153–163. MacWhinney, B. 1975. Pragmatic patterns in child syntax. Stanford Papers And Reports on Child Language Development 10: 153–165. MacWhinney, B. 1977. Starting points. Language 53: 152–168. MacWhinney, B. 1978. The acquisition of morphophonology. Monographs of the Society for Research in Child Development 43, Whole no. 1, 1–123. MacWhinney, B. 1982. Basic syntactic processes. In Language Acquisition, Vol. 1. Syntax and Semantics S. Kuczaj (Ed.), 73–136. Hillsdale NJ: Lawrence Erlbaum. MacWhinney, B. 1987a. The competition model. In Mechanisms of Language Acquisition, B. MacWhinney (Ed.), 249–308. Hillsdale NJ: Lawrence Erlbaum. MacWhinney, B. 1987b. Toward a psycholinguistically plausible parser. In Proceedings of the Eastern States Conference on Linguistics, S. Thomason (Ed.), Columbus OH: Ohio State University. MacWhinney, B. 1989. Competition and lexical categorization. In Linguistic Categorization, R. Corrigan, F. Eckman & M. Noonan (Eds), 195–242. Amsterdam: John Benjamins. MacWhinney, B. 2000. The CHILDES Project: Tools for Analyzing Talk, 3rd Edn. Mahwah NJ: Lawrence Erlbaum Associates. MacWhinney, B. 2005. Item-based constructions and the logical problem. ACL 2005, 46–54. MacWhinney, B. 2008a. Cognitive precursors to language. In The Evolution of Communicative Flexibility, K. Oller & U. Griebel (Eds), 193–214. Cambridge MA: The MIT Press. MacWhinney, B. 2008b. How mental models encode embodied linguistic perspectives. In Embodiment, Ego-Space, and Action, R. Klatzky, B. MacWhinney & M. Behrmann (Eds), 369–410. Mahwah NJ: Lawrence Erlbaum. MacWhinney, B., Bates, E. & Kliegl, R. 1984. Cue validity and sentence interpretation in English, German, and Italian. Journal of Verbal Learning and Verbal Behavior 23: 127–150. MacWhinney, B. & Leinbach, J. 1991. Implementations are not conceptualizations: Revising the verb learning model. Cognition 29: 121–157. MacWhinney, B. & Pléh, C. 1988. The processing of restrictive relative clauses in Hungarian. Cognition 29: 95–141. Marcus, G. 2000. Pabiku and Ga Ti Ga: Two mechanisms infants use to learn about the world. Current Directions in Psychological Science 9:145–147. Meltzoff, A.N. & Decety, J. 2003. What imitation tells us about social cognition: A rapprochement between developmental psychology and cognitive neuroscience. Philosophical Transactions of the Royal Society of London B, 358, 491–500.

 Brian MacWhinney Menzel, R. & Giurfa, M. 2001. Cognitive architecture of a mini-brain: The honeybee. Trends in Cognitive Sciences 5: 62–71. O’Grady, W. 2005. Syntactic Carpentry. Mahwah NJ: Lawrence Erlbaum Associates. Oller, D.K. 2000. The Emergence of the Speech Capacity. Mahwah NJ: Lawrence Erlbaum Associates. Pelphrey, K.A., Mitchell, T.V., McKeown, M.J., Goldstein, J., Allison, T. & McCarthy, G. 2003. Brain activity evoked by the perception of human walking: Controlling for meaningful coherent motion. Journal of Neuroscience 23: 6819–6825. Pelphrey, K.A., Morris, J.P. & McCarthy, G. 2005. Neural basis of eye gaze processing deficits in autism. Brain 128: 1038–1048. Piaget, J. 1954. The Construction of Reality in the Child. New York NY: Basic Books. Pinker, S. 1997. How the Mind Works. New York NY: W.W. Norton & Company. Rizzolatti, G., Fadiga, L., Gallese, V. & Fogassi, L. 1996. Premotor cortex and the recognition of motor actions. Cognitive Brain Research 3: 131–141. Sagae, K., Davis, E., Lavie, E., MacWhinney, B. & Wintner, S. 2007. High-accuracy annotation and parsing of CHILDES transcripts. In Proceedings of the 45th Meeting of the Association for Computational Linguistics. Prague: ACL. Schütz-Bosbach, S. & Prinz, W. 2007. Perceptual resonance: Action-induced modulation of perception. Trends in Cognitive Sciences 11: 349–355. Tomalin, M. 2007. Reconsidering recursion in syntactic theory. Lingua 117: 1784–1800. Tomasello, M., Call, J. & Gluckman, A. 1997. Comprehension of novel communicative signs by apes and human children. Child Development 68: 1067–1080. Tucker, et al. (in this volume). Neural mechanisms of recursive processing in cognitive and linguistic complexity. Westermann, G. & Miranda, E.R. 2004. A new model of sensorimotor coupling in the development of speech. Brain and Language 89: 393–400. Wilson, M., & Knoblich, G. 2006. The case for motor involvement in perceiving conspecifics. Psychological Bulletin 131: 460–473.

Cognitive and neural underpinnings of syntactic complexity Diego Fernandez-Duque Villanova University

Based on a review of the neuroimaging literature, I argue that the resources allocated for processing syntactically complex sentences (i.e., object-extracted relative clauses) are domain-general. Overlapping brain areas are activated by OR clauses and by effortful executive tasks such as storing information in verbal working memory, resolving conflict among competing representations, and switching one’s mindset. A re-conceptualization of ‘syntactic complexity’ in terms of executive functions provides a useful framework in which to explore its links to relational complexity and to cognitive neuroscience. As such, this approach should prove useful to linguists and cognitive scientists alike.

1. Introduction For almost half a century, researchers in psycholinguistics have been interested in the relation between syntactic processing of sentences and domain-general cognitive processes (Miller & Chomsky, 1963). The question was initially addressed on linguistic and philosophical grounds (Fodor, 1988) and tackled soon thereafter by behavioral and neuropsychological experiments (Caplan & Waters, 1999; Miyake, Carpenter, & Just, 1994). Over the last two decades, the emergence of neuroimaging techniques has provided a wealth of information about the relation between language processing and domain-general resources (Grodzinsky & Friederici, 2006; Kaan & Swaab, 2002). A comprehensive review of this literature is beyond the scope of the current paper. Instead, I will focus on the processing of relative clauses. It has been argued that some embedded sentences are more taxing to process than others. In particular, it is often claimed that people have a harder time understanding object-extracted relative (OR) clauses than understanding subject-extracted relative (SR) clauses, as demonstrated by the following example: OR: The reporter that the senator attacked admitted the mistake SR: The reporter that attacked the senator admitted the mistake

 Diego Fernandez-Duque

OR clauses have a non-canonical word order (object-verb-subject), requiring listeners to reorganize the sequential order of a sentence so that it matches its syntax. In these clauses, the perceptual location of the critical phrase is non-adjacent to its semantic interpretation. Therefore, its processing requires a syntactic movement (aka grammatical transformation) across another element. Some researchers have argued that this additional syntactic operation is at the core of what makes OR clauses more difficult to process. A prominent theory of this kind is the Trace-Deletion Hypothesis (Grodzinsky, 2006). Proponents of this view further argue that syntactic movement has its biological substrate in the left frontal cortex, a claim consistent with some aphasia and neuroimaging data (Ben-Shachar, Hendler, Kahn, Ben-Bashat, & Grodzinsky, 2003). In contrast, other researchers have explained the increased processing cost of OR clauses by appealing to ‘syntactic complexity.’ Syntactic complexity theories differ from each other regarding how complexity is defined. Some are explicit in their definition and provide a metric of complexity that can be use to test predictions (Gibson, 1998; Halford, Wilson, & Phillips, 1998). Others leave the concept of complexity undefined and focus instead on the experimental conditions that lead to increased complexity (Caplan, Alpert, & Waters, 1998; Carpenter, Miyake, & Just, 1994; Friederici, Fiebach, Schlesewsky, Bornkessel, & von Cramon, 2006). Theories of syntactic complexity also vary from each other on how general are the resources used in syntactic processing. Some theories posit the existence of a limited-capacity memory devoted exclusively to the processing of syntactic relations (Caplan et al., 1998; R. Lewis, 1996). Others argue that syntactic complexity taps onto cognitive resources that are shared with other non-linguistic complex tasks (Andrews, Birney, & Halford, 2006; Larkin & Burns, 1977; Miyake et al., 1994). According to syntactic complexity theories, OR clauses are said to be more difficult to process than SR clauses for many reasons, including the following ones: 1. OR clauses pose a larger storage cost than SR clauses. This is because partially analyzed clauses need to be stored in short-term memory until their completions are available. In other words, the first noun phrase (the reporter) has to be retained in working memory until the verb (attacked) is encountered, at which point syntactic and thematic integration can occur. Once the information is integrated, it becomes part of the text meaning and of the long-term memory representation of that sentence. However, until it is integrated it needs to be held in working memory. 2. OR clauses yield more syntactic ambiguity, as more than one syntactic structure is applicable at the beginning of the sentence. For example, instead of “the reporter the senator attacked admitted the error” one could say “the reporter the senator …and the president disagreed.” Comprehension is improved when ambiguity is

Cognitive and neural underpinnings of syntactic complexity 

eliminated by adding the pronoun ‘whom’, as in “the reporter whom the senator…” (Hakes & Foss, 1970). 3. OR clauses pose a larger demand for syntactic integration. In OR clauses there is a longer distance between dependents (reporter, attacked) and as additional words are processed, the activation level of the initial element decays.1 Therefore more resources are required for the reactivation of the initial element at the time of integration. 4. OR clauses require perspective shifts and therefore pose a larger thematic integration cost (MacWhinney & Pleh, 1988). In sentences with OR clauses, the first noun plays two different thematic roles. In our example, ‘the reporter’ starts as the subject of the main clause, shifts to being the object of the attack in the relative clause, and goes back to being the subject ‘who admitted the error’ at the end of the sentence. Such perspective shifts during sentence processing mean that the two competing representations have to be coordinated. Comprehension is enhanced when the first noun is an inanimate object, as in “the rock the kid touched was hot” (Hakes, 1972). In such cases, there is no bias toward interpreting the initial noun as a subject; therefore switching to an ‘object’ representation becomes easier. It may be useful to compare the processing of syntactic complexity to Executive Function processes, which play a prominent role in theories of working memory, consciousness, and willful action. In most of these theories, one of the properties of executive functioning is its domain-generality. Although the taxonomy of executive function is itself a matter of controversy, executive functions are thought to contribute to: a. manipulating representations in working memory, as opposed to merely storing them. b. coordinating ambiguous or conflicting information. c. switching one’s mindset to facilitate the interpretation or implementation of new rules for guiding behavior.

1. According to some theories of syntactic complexity, the distance between dependents is determined not so much by the total number of words to be processed before integration, but rather by the number of new discourse structures (Gibson, 1998). Each time a new discourse referent occurs, cognitive resources have to be deployed to include it in the discourse environment. In contrast, when the new referent is already part of the discourse environment (e.g., indexal pronouns), OR clauses become easy to process (e.g., the book you bought…).

 Diego Fernandez-Duque

From the point of view of a cognitive neuroscientist who specializes in executive function but knows much less about psycholinguistics, the similarities between syntactic complexity of relative clauses and executive function seem, at first sight, quite compelling. On the other hand, these similarities may stem from comparing two very broad and illdefined concepts, rather than from a genuine conceptual overlap. Thus, the remainder of this chapter is devoted to comparing the properties of syntactic complexity of relative clauses and executive function, in an attempt to systematically uncover their possible relation. For this, I will analyze the anatomical overlap between processing syntactic complexity and executive function tasks, with a focus on the frontal cortex. I will also explore the conceptual similarities between tasks that tap syntactic complexity and tasks that tap cognitive complexity in non-syntactic domains.

2. Neuroimaging of syntactic complexity This is not an exhaustive review of the neurology of syntactic complexity but rather a review more limited in scope. From an anatomical standpoint, this review focuses on the inferior gyrus of the frontal cortex, an area that has been implicated in both the processing of syntactic complexity and the processing of many executive function tasks. As a measure of syntactic complexity, the review focuses on the comparison between OR and SR clauses. From a methodological standpoint, the review focuses on neuroimaging research (fMRI, ERP) with only a brief mention of neuropsychological data.

2.1 N eural activation in OR vs. SR processing, and the influence of non-syntactic factors To assess the neural substrates of syntactic complexity, neuroimaging studies have often compared object-extracted and subject-extracted relative clauses. This comparison has consistently shown left hemisphere activation of frontal and temporal areas (Caplan et al., 1998; Just, Carpenter, Keller, Eddy, & Thulborn, 1996). In the frontal lobe, the activation is centered in the inferior-frontal gyrus (IFG) particularly in its pars opercularis (Brodmann’s Area 44) and its pars triangularis (BA 45). Sometimes these areas are referred to as Broca’s area, but I will refrain from using this label, as its anatomical boundaries remain elusive and its use is bound to create confusion (Lindenberg, Fangerau, & Seitz, 2007).2 Activation of IFG is consistent with

2. Many studies refer to ‘Broca’s area’ without a clear definition of its anatomical boundaries. Even in studies that do define the boundaries, there is significant variability on what those boundaries are. Part of the reason is that the macroscopic features are not reliable landmarks for its cytoarchitectonic borders (Amunts et al., 1999). To further complicate matters, lesions to ‘Broca’s area’ are neither necessary nor sufficient for the syntactic deficits observed in Broca’s aphasia (Dick et al., 2001).

Cognitive and neural underpinnings of syntactic complexity 

neuropsychological evidence showing that lesion to IFG area reduces comprehension of embedded clauses with non-canonical word order (Friederici, 2002).

44 6 45 47

22 37 22 21

OR clauses differ from SR clauses in terms of syntactic operations, as the canonical word order needs to be reconstructed in the former but not the latter. Thus, one can be tempted to attribute the IFG activation to the extra syntactic processing that OR clauses demand. However, OR and SR clauses also differ in other respects such as the amount of effort involved in information processing. In other words, syntax complexity is confounded with non-syntactic cognitive processes, such as working memory. To disentangle the contribution of syntax and non-syntactic processes, a useful approach is to include a ‘cognitive’ factor to the experimental design, and explore whether it interacts with syntactic complexity. The assumption is that computations carried out by overlapping neural substrates will interact. Therefore, the cognitive factor included in the design is usually one known to elicit IFG activation. Reading low-frequency words causes more activation of left IFG than reading high frequency words.3 This raises the question of whether the effect of syntactic complexity in IFG will be modulated by word-frequency. To address this question, OR and SR sentences with high-frequency or low-frequency words were created in a factorial design (Keller, Carpenter, & Just, 2001). As in previous studies, left IFG showed greater activation for the processing of OR clauses than for the processing 3. Left IFG activation by low frequency-words may be due to lexical selection, as lowfrequency words demand more intense filtering of distracting lures (Badre & Wagner, 2007; Thompson-Schill, D’Esposito, Aguirre, & Farah, 1997). Increased IFG activation by low frequency words may also stem from phonological recoding. For example, IFG is activated by reading pseudo-words (i.e., English-like words absent of meaning), a task that requires grapheme-to-phoneme conversion but does not require lexical retrieval. There might even be some amount of anatomical segregation in parts of the IFG between these two mechanisms (Fiebach, Friederici, Muller, & von Cramon, 2002; Poldrack et al., 1999).

 Diego Fernandez-Duque

of SR clauses, and greater activation for low-frequency words than for sentences with high-frequency words. More importantly, there was an interaction between these two main effects: the effect of syntactic complexity on IFG was evident only for sentences of low-frequency words. These findings point to a common anatomical substrate for the processing of syntactic complexity and the processing of nonsyntactic operations such as phonological recoding and/or lexical selection. Other fMRI studies have used the factorial design to provide evidence that IFG is not the locus of syntactic movement per se, but rather it supports aspects of working memory. In one such study, syntactic complexity was varied independently from working memory in German indirect wh- questions (Fiebach, Schlesewsky, Lohmann, von Cramon, & Friederici, 2005). Object-initial questions were compared to subjectinitial questions with the same working memory demands. This comparison did not reveal IFG activation. A different set of results was obtained for sentences with varying working memory load. In some of these sentences the verb was dislocated from its canonical position over a relatively long distance while in others the dislocation was shorter. These two types of sentences had different working memory load. Those with larger working memory load did cause larger IFG activation (BA 44). In a study of OR vs. SR clause processing, whether the disambiguation occurred early or late within a clause was systematically varied (Fiebach, Vos, & Friederici, 2004). This early-versus-late manipulation was intended as a manipulation of working memory load. Based on a different measure of working memory capacity, subjects were classified as having high- or low- working memory span. OR clauses caused greater IFG activation than SR clauses, consistent with previous findings. More importantly, the effect of syntactic complexity was dependent on the working memory demands. More specifically, increased IFG activation by syntactically complex sentences was evident only for participants in the low WM span group while reading sentences that demanded most WM (i.e., in which disambiguation occurred late in the clause). In other words, it seems that working memory demands were the main cause of IFG activation. In sum, there is substantial evidence to argue that IFG is not recruited exclusively for the syntactic reconstruction of canonical word order but rather is implicated in working memory or processing load. As just described, IFG activation by syntactic complexity is modulated by sentence ambiguity, lexical retrieval, and other memory demands.

2.2 Mechanisms underlying IFG activation in OR clause processing In subject-relative clauses, syntax helps to integrate nouns with verbs: once the information is integrated, it becomes part of the long-term memory representation of that sentence. In contrast, in the object-relative clause the partially processed

Cognitive and neural underpinnings of syntactic complexity 

but incomplete syntactic dependencies need to be maintain in working memory. Based on this analysis, working memory differences should begin with the occurrence of the second noun-phrase in the object-relative clause (e.g., ‘the reporter that the senator attacked admitted the mistake’). On the other hand, it is at the end of the object-extracted relative clause that syntactic and thematic integrations occur. The verb of the main clause resolves the ambiguity and allows assigning the roles of ‘who did what to whom’. If the cost of OR clause processing stems from these syntactic and thematic integrations, the IFG activation should start near the end of the OR clause. These two hypotheses are not necessarily incompatible: it is possible that the IFG activation is driven both by working memory demands and integration costs. Unfortunately, fMRI studies are unable to assess these alternative hypotheses because in fMRI studies the hemodynamic response lags the neuronal response by several seconds, making it impossible to test which part of the sentence is triggering the activation. This limitation can be overcome using online measures such as gaze duration (Holmes & O’Regan, 1981; Traxler, Williams, Blozis, & Morris, 2005), word-byword reading (Gibson, Desmet, Grodner, Watson, & Ko, 2005), and pupil diameter (Just & Carpenter, 1993). Studies using these methodologies indicate that the point of greatest effort is at the end of the object-relative clause, when thematic roles are assigned (i.e., when it is decided who did what to whom). Although useful as online measures of performance, these methodologies do not allow direct comparisons to brain activation. This limitation is overcome by event-related scalp potentials (ERPs), as online electrophysiological measures of sentence processing can be correlated with fMRI activation. Although ERPs’ spatial resolution is not as good as that of fMRI, its temporal resolution is much superior. Thus, the methodological strengths of the two techniques complement each other very well. ERP studies provide support for the working memory hypothesis. The electrophysiological response to the OR clause begins to diverge from the SR clause at the appearance of the second noun (the senator) which marks the beginning of a differential working memory load between OR and SR sentences (King & Kutas, 1995). This divergence occurs in left anterior sites (i.e., frontal lobe) and is similar to the effect found when working memory load is increased in other types of sentences (Kluender & Kutas, 1993). Furthermore, the left anterior negativity is also found when comparing SR clauses to unembedded sentences, consistent with the increased memory demands of embedded sentences. Interestingly, the laterality of the OR effect occurs only for reading material; auditory presentation elicits a bilateral effect instead (Muller, King, & Kutas, 1997). Thus it seems likely that phonological recoding may contribute as a modulator of this effect. Finally, ERP studies also support the integration hypothesis. In fact, the largest ERP difference between OR and SR does occur at the end of the OR clause, when the main clause verb is first displayed (King & Kutas, 1995).

 Diego Fernandez-Duque

2.3 Some evidence of syntactic specificity in left IFG activation According to the literature reviewed so far, IFG activation while processing OR clauses is mediated by non-linguistic cognitive processes such as working memory and perspective taking. However, some studies argue otherwise. One fMRI study assessed activation under different levels of (a) syntax complexity and (b) speech rate (Peelle, McMillan, Moore, Grossman, & Wingfield, 2004). It compared OR clauses to SR clauses; speech speed was systematically manipulated. Syntax complexity activated left IFG across all presentation rates. Fast presentation rate elicited medio-frontal activation usually activated by effortful tasks. More importantly, speech rate did not modulate the level of activation due to syntax complexity. This lack of interaction is at odds with the behavioral data in the same task showing that the cost of increased syntax complexity is modulated by speech presentation rate, with larger error rates for OR sentences at faster speech rates. A second study showed activation in left frontal cortex independent of a variety of factors that were manipulated to increase demands (Ben-Shachar et al., 2003). These results have sometimes been interpreted as evidence for a core network of brain regions that supports grammatical processes and includes IFG and postero-lateral temporal cortex (Cooke et al., 2006). Additional brain regions are thought to be engaged as required by extra cognitive demands. It is unclear how best to reconcile these findings and those showing interaction.

3. Neuroimaging of executive function ‘Executive Function’ is an umbrella term for a wide range of functions that contribute to working memory, consciousness, and willful action. A central goal in cognitive science has been to describe how those functions relate to each other and to other cognitive systems (Miyake, Friedman, Emerson, Witzki, & Howerter, 2000). Cognitive neuroscience has joined the enterprise by exploring whether the same brain areas, most notably in the frontal cortex, are recruited for different aspects of executive control. Although there is not yet a definitive taxonomy of executive function, there is general agreement that a central place should be given to the following abilities: a. manipulating mental representations in working memory b. coordinating ambiguous or conflicting information c. switching mental sets Some of the experimental paradigms developed over the years aim to explore a single executive function and fractionate it into its more basic subcomponents. Other paradigms aim to relate executive function to other cognitive systems, such as working memory (Baddeley, 1992) and visuospatial attention (Fernandez-Duque & Posner, 2001).

Cognitive and neural underpinnings of syntactic complexity 

In such latter cases, ‘executive function’ is conceptualized as a component of the cognitive system in question. Finally, executive processes are sometimes involved in tasks designed to study some other function. For example, the ability to ignore distractors is an executive function that modulates performance in lexical decision tasks. Given the central role that verbal working memory seems to play in syntactic complexity, I start by reviewing executive function as part of working memory capacity. This requires a brief description of Working Memory (WM) as a system that allows people to actively maintain and manipulate information. One of most influential models of WM is the one proposed by Baddeley (Baddeley, 1992). That model poses the existence of a system for maintaining verbal information known as the phonological loop, a system for maintaining visual information, and a central executive system for manipulating the information.

3.1 Simple storage of verbal information in working memory For functioning in everyday life, it is absolutely necessary to be able to maintain information after it ceases to be perceptually available. Otherwise, we would be unable to hold a phone number in mind or to understand any sentence more than a few words long. This simple storage of information is dependent on the phonological loop. Presumably, this is the aspect of WM that is tapped by lengthening the distance between syntactic dependencies. Neuropsychological and neuroimaging evidence shows that the phonological loop is lateralized to the left hemisphere and is further subdivided into a subvocal rehearsal process and a passive storage of phonological information. A recent meta-analysis has confirmed that simple storage of verbal information in working memory activates IFG and that such activation is lateralized to the left hemisphere (Wager & Smith, 2003). Left IFG is also active during rhyming judgment tasks and other phonological tasks, consistent with a close functional relation between verbal working memory and silent speech (Paulesu, Frith, & Frackowiak, 1993; Poldrack et al., 1999). This raises the possibility that increased left IFG activation in response to OR clauses may be due to increased phonological rehearsal. According to this hypothesis, increased complexity would bias participants toward sounding their words out. If this hypothesis is correct, the left IFG activation by OR clauses should disappear under conditions that prevent silent speech (i.e., articulatory suppression). However, some of the evidence suggests otherwise: left IFG activation by OR clauses occurs even when participants read the sentences while uttering an unrelated word every second, aimed at suppressing silent speech (Caplan, Alpert, Waters, & Olivieri, 2000). Nonetheless, there is some neuropsychological evidence in favor of the hypothesis. For example, comprehension of OR clauses is impaired in patients whose clinical symptoms include effortful speech and dysarthria. One such example is patients with progressive

 Diego Fernandez-Duque

non-fluent aphasia, a type of dementia with brain atrophy most pronounced in left lateral frontal cortex (Grossman & Moore, 2005).

3.2 Manipulating information in verbal working memory As suggested by the label working memory, people are able not only to store information but also to manipulate and reorganize it. Such ability correlates with individual differences in reasoning, planning, and other intelligent behavior (Kane & Engle, 2002). The ability to manipulate information in WM has been tested with several different paradigms. In the n-back task, letters are presented one at time separated from each other by a delay of 2 seconds. For each letter, the participant has to decide whether it matches the letter presented n stimuli back. As an example, imagine that in a 2-back task you see the following letter sequence: G, T, L, B, L. Upon seeing the first L, you should report it does not match the reference letter G, that is, the letter that occurred two trials back. As soon as this decision is made, you have to update the information, replacing G for T as the reference letter. Although some studies report IFG activation in the 2-back task, it seems likely that such activation is due to increased verbal rehearsal rather than to working memory updating. In support of this interpretation, silent rehearsal tasks cause as much left IFG as 2-back tasks (Awh et al., 1996). This is consistent with the findings from a recent meta-analysis showing that working memory updating does not increase left IFG activation relative to the activation in the simple storage condition (Wager & Smith, 2003). In the few studies in which updating does activate IFG, the effect is lateralized to the right hemisphere. In sum, it seems that left IFG activation is due to rehearsal rather than updating, at least as far as the n-back task is concerned. In the alphabetization task, a sequence of letters is presented followed by a delay. During the delay, the participant has to organize the letters in alphabetical order. When the probe appears, the subject reports its location in the alphabetical order. The alphabetization condition is compared to a storage condition in which the letters have to be retained in the order they are presented. Both conditions equally activate IFG, consistent with a rehearsal interpretation. The alphabetization condition uniquely activates more dorsal areas of the frontal cortex, such as Brodmann’s Areas 9 and 46. Thus, working memory manipulation appears to depend on neighboring areas of IFG, rather than IFG proper (D’Esposito, Postle, Ballard, & Lease, 1999). The Operation Span task assesses the ability to maintain words in memory while solving math problems. In this dual-task paradigm, a sequence of words is presented one at a time, each paired with a math equation. The task requires holding certain information in mind while doing something else. Performance in the Operation Span task correlates with individual differences in reasoning and general fluid intelligence

Cognitive and neural underpinnings of syntactic complexity 

(Engle, Tuholski, Laughlin, & Conway, 1999). Activation is more anterior than for syntactic complexity tasks and lateralized to the right hemisphere. It includes Brodmann’s Areas 10, 46, and 47 (Wager & Smith, 2003).

3.3 Summary of WM activation and its relation to syntactic complexity The literature provides support for a common anatomical substrate for syntactic complexity and some aspects of working memory, but not others. On the one hand, there is firm evidence in favor of left IFG activation for subvocal rehearsal and for syntactic complexity. This is consistent with the view that OR clauses require maintaining more information in working memory for longer time. On the other hand, there is less evidence of overlap between syntactic complexity and manipulation of information in working memory. Although syntax complexity and WM manipulation both activate lateral frontal cortex, the precise areas of activation are mostly non-overlapping. However, it remains a possibility that overlap exists with other executive functions. I discuss this possibility next.

3.4 Coordinating ambiguous or conflicting information Objects in the environment have many attributes; they have color, shape, motion, size, function, location in space, meaning, etc. For any given task, only a few of those attributes are relevant. Effective information processing requires, among other things, the ability to adequately select which information to process and which to ignore. For example, when picking an apple it is useful to know its size and location, but it is not necessary to know its color. However, color information may become useful when deciding whether to eat the apple, as color will cue the eater to any rotten parts as well as to the flavor she should expect. The above example illustrates some of the difficulties confronting an agent as she processes a multi-attribute stimulus. On the one hand, some of the perceptually salient attributes carry information that should be ignored because they are useless or even harmful to performance. On the other hand, those same attributes may become relevant at some later point in time. This variability requires that the agent be flexible when deciding how to allocate her attention. In cognitive psychology, the filtering of salient information has been studied under the banner of conflict resolution. The flexible allocation of attention to different dimension has been studied under the banner of set switching.4

4. Sometimes both terms are referred to as ‘selective attention’, a label that highlights that these processes are voluntary, effortful, and require the deployment of domain-general resources.

 Diego Fernandez-Duque

3.4.1 Conflict resolution in perceptual tasks The prime example of conflict resolution is the Stroop task, in which subjects are instructed to respond based on a certain stimulus dimension (e.g., hue) while ignoring some other information (e.g., word meaning). When information from the distracting dimension is incongruent with the target dimension (e.g., the word RED in green ink), conflict arises (Bush, Luu, & Posner, 2000). Its resolution depends on subjects’ ability to ignore the irrelevant information, which in turn requires keeping in mind the correct mindset (i.e., ‘respond to hue’). This version of the task is sometimes referred to as the ‘verbal’ Stroop. Other Stroop-like tasks require the filtering of non-verbal information. The spatial-compatibility task, for example, requires participants to ignore the stimulus location and respond instead based on stimulus shape. The ‘flanker task’ requires responding to the center target while ignoring distractors on the sides (Fan, Flombaum, McCandliss, Thomas, & Posner, 2003). Performance in the verbal Stroop task is impaired following lesion to left IFG (Hamilton & Martin, 2005) and neuroimaging studies show left lateralized IFG activation for incongruent trials (RED in green ink) (Derrfuss, Brass, Neumann, & von Cramon, 2005). The left lateralization of these effects is specific to verbal material: conflict in a non-verbal flanker task activates right IFG (Hazeltine, Poldrack, & Gabrieli, 2000), and lesion to left IFG does not impair performance in the non-verbal spatial-compatibility task (Hamilton & Martin, 2005).

3.4.2 Conflict resolution in working memory tasks In the Stroop and Stroop-like tasks, the filtered information is perceptually available. In contrast, other tasks require the filtering or inhibition of memory representations. Such tasks are relevant to our discussion of syntactic complexity because interpreting a sentence based on new information does require suppressing the no longer relevant interpretation. One task requiring inhibition of memory representation is the proactive interference task. In this task, a few words (or letters) are presented one per second, immediately followed by a probe. The participant reports whether the probe matches one of the items she just saw. No-match trials can be further divided based on probe familiarity: a probe is said to be familiar if it appeared as an item in the preceding trial. Familiar probes in non-match trials lead to slower and less accurate responses. In those trials, there is conflict between the familiarity of the probe and its absence in the short-term memory set. High conflict trials activate left IFG (Brodmann’s area 45); this activation is triggered by the probe onset, which suggests that it is related to conflict resolution (Postle, Brush, & Nick, 2004). Consistent with this interpretation, performance in high conflict trials is impaired following left IFG lesion (Hamilton & Martin, 2005). Impairment can also be triggered in healthy adults by temporally

Cognitive and neural underpinnings of syntactic complexity 

inactivating left IFG with the use of repeated transcranial magnetic stimulation (Feredoes, Tononi, & Postle, 2006).

3.4.3 Conflict resolution in semantic memory tasks When trying to retrieve a word from memory, it is necessary to filter out distractors semantically related to the target. For example, imagine that you are shown a picture of a pencil and asked to name its function. The first word that will come to mind is ‘pencil’, rather than the correct answer. Or imagine you are shown a picture of an ox and asked to name which animal it is. In this case, a more prototypical member of the category — e.g.,‘cow’- may to come to mind. As these examples illustrate, correct performance in these tasks require suppressing the tendency to use the most salient response. Neuropsychological and neuroimaging studies indicate that left IFG plays an important role in such inhibition (Kan & Thompson-Schill, 2004).

3.5 Switching mental sets When processing multidimensional stimuli, attention is allocated selectively to a particular dimension. However, the focus of attention can be shifted voluntarily to a different stimulus dimension, or even to a different task. For example, a person who had previously been responding based on stimulus color (red, blue) can begin to respond based on stimulus shape (triangle, square). Switching requires selecting the new mental set (‘respond to shape’) and inhibiting the old one (‘respond to color’). It also requires the activation of specific rules (‘if red, press left’). These two components of set switching are dissociable in the brain: Set selection activates lateral and medial prefrontal cortex, while rule activation activates mostly the intra-parietal sulcus (Derrfuss et al., 2005; Wager, Jonides, & Smith, 2006). A recent meta-analysis found that set switching tasks and verbal Stroop tasks cause overlapping activation in the posterior part of the left IFG (Derrfuss, Brass, & von Cramon, 2004). In switch trials, this area — known as the inferior frontal junction — becomes active even before the appearance of the target (Brass & von Cramon, 2002). This is consistent with a role in set selection, as behavioral studies have found that rule activation requires a perceptually available target (Monsell, 2003). Furthermore, neuroimaging studies show activation of this brain area at the start of a block of trials, once participants are instructed to get ready for the task (i.e., to adopt the appropriate mindset) (Dosenbach et al., 2006). The set selection is likely to involve the maintenance of task-relevant information, probably in verbal format. Such maintenance of task-relevant information is a process akin to selective attention, the mental highlighting of some stimulus property for the benefit of preferred processing and conscious awareness.

 Diego Fernandez-Duque

One fMRI study provides striking evidence that the left lateralization of set switching is related to verbal processing. In this study, overlapping face/word stimuli were displayed and participants either performed a gender task on the face (male/female) or a syllable-counting task on the word (two syllables or not). Every four trials, a cue signaled participants to continue the same task or switch to the other task. Left IFG was activated when performing the word task and right IFG was activated when performing the face task. More importantly, those effects were larger for switch trials than for repeat trials (Yeung, Nystrom, Aronson, & Cohen, 2006). In other words, the IFG activation was larger for the trials that required more attention. Neuropsychological studies have further shown that lesions in the left hemisphere impair performance in switch trials (Mayr, Diedrichsen, Ivry, & Keele, 2006). Besides causing increased local switch cost (the cost in switch trials), left lateral frontal lesions also lead to increase global switch costs. Global switch costs refer to slow responses to no-switch trials in blocks with bi-dimensional stimuli, relative to blocks with unidimensional stimuli. When the relevance of each dimension alternates every few trials, the irrelevant dimension becomes salient, and more attention is needed to filter it out. If this explanation is correct, we should find that subjects who are least effective at set selection are more exposed to the irrelevant dimension and therefore have more conflict to resolve. Neuroimaging studies in normal subjects support this prediction: participants who perform poorly in switch trials show increased activation in some conflict resolution areas of the frontal lobe (superior and middle frontal gyri) (Wager et al., 2006). Besides being the source of selective attention in some verbal switching tasks, parts of left IFG are also the target of selective attention. For example, when participants have to attend to syntax (e.g., plausibility judgment task), the processing of OR clauses trigger larger IFG activation than when syntax is not task relevant (e.g., detecting the presence of a pseudo-word) (Chen, West, Waters, & Caplan, 2006).

4. C oordinating information in relative clauses: A role for conflict resolution and set switching In explaining differences between OR and SR clauses, studies of syntactic complexity often appeal to concepts such as ‘working memory’ or ‘cognitive load’. The strength of those explanations lies on a detailed description of what the terms mean and how they relate to syntactic processing. This has been done with some success for some aspects of working memory (Caplan & Waters, 1999). In contrast, less is known about the conditions under which conflict resolution and set switching would contribute to syntactic processing (Novick, Trueswell, & Thompson Schill, 2005). Some likely candidates are mentioned next. Conflict resolution and switching of mindset are likely more involved in the processing of OR clauses than in the processing of SR clauses. One reason for this is that

Cognitive and neural underpinnings of syntactic complexity 

OR clauses are more ambiguous: in OR clauses, the lexical semantics often do not converge with the default syntactic order. Consider the following two sentences: a. the girl the boy kicked was big. b. the ball the boy kicked was big. These are two OR clause sentences, but ‘a’ is harder to understand than sentence ‘b’. The reason for this is that ‘girl’ is an animate noun (Mak, Vonk, & Schriefers, 2002). Based on semantic knowledge, the reader builds the expectation that the animate noun will be the agent of the action. Furthermore, in the English language it is very infrequent for an animate noun to be followed by an OR clause (Fox & Thompson, 1990). Thus, the expectation based on distributional statistics agrees with the expectation based on lexical semantics, namely that following an animate noun the clause will be a SR. In OR clauses, this expectation needs to be overcome, a process that requires conflict resolution. Consistent with this account, OR sentences in which the first noun is animate elicit greater left IFG activation than OR sentences in which the first noun is inanimate (such as ‘b’ above) (Caplan, Chen, & Waters, in press). It is likely that the activation is elicited at the end of the relative clause. OR sentences require the rapid processing of adjacent verbs, increasing the chances that role assignments for the two clauses will overlap in time. Such temporal overlap is bound to create conflict, as the same noun has to be assigned two different roles (patient of the relative clause, agent of the main clause). This conflict will be greater in the OR sentence with an animate noun in the main clause because in that case there is a stronger expectation that is being violated. ERPs using a word-by-word reading paradigm reveal that the animacy effect is elicited in part by the relative clause verb, and in part by the main clause verb (Weckerly & Kutas, 1999). The relative clause verb elicits an effect that is centered in centroparietal sites, with a scalp topography and time course similar to the P600 component. It is probably the neural marker of the mismatch between the incoming stimulus and the current mental set, or possibly a marker of working memory update (Coulson, King, & Kutas, 1998).5 The second ERP component is elicited by the main clause verb.

5. The functional significance of the P600 is a matter of debate. Some researchers argue that it is a general-purpose process related to the updating of information in working memory. In support of this view, the P600 has a scalp distribution similar to that of another ERP component associated with memory updating, the P3b (Coulson et al., 1998). Others argue that the P600 is caused by the syntactic reanalysis that occurs whenever the parser fails to find a meaningful parse (Friederici, 2002). Consistent with this interpretation, the P600 is triggered by syntactic violation and by sentences with correct but non-preferred structure, such as OR sentences and garden-path sentences.

 Diego Fernandez-Duque

It is a left anterior negativity (LAN) that occurs in the 200–500 ms window after stimulus onset. Its location is consistent with IFG activation. Interestingly, the LAN effect is obtained only in participants whose sentence comprehension is good. It would be interesting to explore whether the effect correlates with conflict resolution, set switching, or both. Support for the claim that conflict resolution is related to syntactic complexity also comes from the observation that OR clauses with indexal pronouns are easy to process. Consider the following two examples: a. The reporter the senator attacked admitted the error. b. The reporter you attacked admitted the error. Sentence (a) is harder to understand than sentence (b). A likely interpretation is that in (a) the embedded subject (the senator) is a new referent in the discourse while in (b) the embedded subject (the pronoun ‘you’) is already part of the discourse, at least implicitly. According to one of the more prominent theories of syntactic complexity, the presence of a new discourse referent makes more difficult the integration (attacked-reporter) because processing new discourse information demands working memory capacity (Gibson, 1998). As a consequence, access to the main clause noun becomes more difficult and so does the noun-verb integration. The difficulty in accessing the noun is probably due to retrieval interference, although other factors may also come into play (R.L. Lewis, Vasishth, & Van Dyke, 2006; Van Dyke & McElree, 2006). To overcome that interference, conflict resolution needs to be applied. Conflict resolution may also explain why processing OR clauses is more difficult when the noun phrases are similar (Gordon, Hendrick, & Johnson, 2001). Consider the following two OR clauses: a. the actor that the director thanked worked in many hit movies before 1980 b. the actor that Fay thanked worked in many hit movies before 1980 Sentence ‘a’ is harder to understand than sentence ‘b’, and participants in a word-byword reading paradigm slow down at the moment of thematic integration (i.e., when processing the verbs of sentence ‘a’). In (a) the embedded noun phrase (the director) is similar to the matrix noun phrase (the actor). In (b) the embedded noun phrase is a proper name (Fay) and thus it is less similar and probably causes less interference. Once again, it is at the moment of retrieval that such interference is likely to occur (R.L. Lewis et al., 2006; Van Dyke & McElree, 2006). Consistent with this view, eye

Cognitive and neural underpinnings of syntactic complexity 

tracking studies show that the similarity effect first appears at the moment of processing the verbs (Gordon, Hendrick, Johnson, & Lee, 2006). Finally, set switching may help explain the pauses that occur during speech production of relative clauses. According to a prominent view, intonational boundaries are the result of resource processing demands on language production (Watson & Gibson, 2006). This ‘cognitive load’ interpretation argues that pauses are due to effortful processing (i.e., executive functions). The claim is that intonational boundaries “provide the speaker with time to (a) plan the properties of upcoming linguistic structure and (b) recover from expending resources after producing complex linguistic structure” (pp. 1045–1046) (Watson, Breen, & Gibson, 2006). According to this cognitive load hypothesis, intonational boundaries should be most likely to occur before and after long constituents because these locations are likely points for planning and recovery (Watson et al., 2006). A further prediction would be that intonational boundaries will be most likely for sentences with large set switching demands. Testing this latter prediction might be difficult, as syntactic complexity and set switching are often confounded. Thus, a goal for future research should be to develop experimental sentence comprehension and speech production designs that vary set switching independently from syntactic complexity. Another approach would be to appeal to individual differences, and test whether intonational boundaries are more likely in participants whose set switching skills are below-average. This approach could also be extended to clinical populations that show impairment in these domains. One such group is Parkinson Disease (PD) patients. PD patients are impaired at set switching in both motor and perceptual tasks (Hayes, Davidson, Keele, & Rafal, 1998). They also have difficulty understanding OR clauses and other syntactically complex sentences. In PD, the deficit in set switching, as well as the deficit in syntactic complexity, is ameliorated by dopamine treatment (Grossman et al., 2001; Hayes et al., 1998). Many patients go on and off medication as part of their treatment, a regimen that offers a rare opportunity to explore the chemical mechanisms underlying syntactic complexity and/or set switching. More generally, PD may prove to be a useful model for testing how executive functions and syntactic complexity interact in the brain. The frontal-striatal loop is a circuit that includes parts of the cerebral cortex as well as sub-cortical structures such as the basal ganglia. This loop is involved in many aspects of behavior, including working memory (Koelsch et al., 2008; McNab & Klingberg, 2008) and set switching (Crinion et al., 2006), as well as learning (Packard & Knowlton, 2002), reasoning (Goel, Gold, Kapur, & Houle, 1997), motor control, and other functions. The hallmark of PD is a dysfunction of this frontal-striatal loop due to partial depletion of the neurotransmitter dopamine. PD deficit in syntactic complexity, such as OR clause comprehension, correlates with deficits in executive tasks such as the Stroop task (Grossman, Lee, Morris,

 Diego Fernandez-Duque

Stern, & Hurtig, 2002) and set switching tasks (Hochstadt, Nakano, Lieberman, & Friedman, 2006).6 The effects of syntactic complexity and working memory maintenance in PD were explored in an fMRI study using a factorial design (Grossman et al., 2003). Type of relative clause (OR, SR) was used to manipulate syntactic complexity, and length of noun-gap linkage (short-distance dependency, long-distance dependency) was used to manipulate memory storage. In healthy adults, both cortical and sub-cortical components of the loop were recruited by syntactically complex sentences with high memory demand. In PD, those sentences only activated the cortical component (left IFG), failing to activate the subcortical component (striatum). These results are consistent with IFG playing a compensatory role to the subcortical dysfunction. It is unclear whether the compensatory effect is on memory or conflict resolution.

5. C onceptual similarities between tasks of syntactic and non-syntactic complexity The reviewed literature on IFG activity shows commonalities between syntactic complexity and some aspects of working memory and executive function. However, to argue that the syntactic and non-syntactic domains are functionally related it is also important to seek commonalities at other levels of analysis. One promising approach is to compare tasks of syntactic complexity to tasks that share a similar structure, or require a similar set of computations. Tasks of reasoning and intelligence seem likely candidates. I discuss those next.

5.1 Transitive inference task Consider the following scenario: Premise 1: Mary is taller than Joan. Premise 2: Joan is taller than Emma. Conclusion: Therefore, Mary is taller than Emma. This type of reasoning is relatively easy. If after reading the two premises you were asked who is tallest, you probably would have no difficulty answering “Mary”. Now, let’s switch the order of the premises:

6. Another factor that may contribute to poor sentence comprehension is PD’s slow information processing, as revealed by abnormally late effects of lexical priming in this clinical group.

Cognitive and neural underpinnings of syntactic complexity 

Premise 2: Mike is taller than Joe. Premise 1: Ed is taller than Mike. Conclusion: Therefore, Ed is taller than Joe. This slight modification makes the second example much more difficult to process. Why? In trying to answer this question, it is worth pointing out that performance in the transitive inference task is correlated with OR clause comprehension (Andrews et al., 2006). Remarkably, the correlation remains significant after controlling for performance in SR clause comprehension, and for performance in a variant of the Operation Span task. This raises the possibility that a common factor underlies performance in OR clause comprehension and transitive inference reasoning. The factor, it has been proposed, is relational complexity or number of related dimensions that need to be considered simultaneously (Andrews et al., 2006). In the transitive reasoning task, the goal is to rank each person by height, based on relative height information. In the first example, the relations between names can be processed sequentially. This segmentation reduces the task complexity: by the time the second relation is being processed, the first one is already solved. In contrast, in the second example both relations need to be considered simultaneously. In other words, the relational complexity of the task is increased, and so its difficulty. The processing of SR and OR clauses lends itself to a similar analysis. In this case, the goal is thematic role assignment (‘who did what to whom’). Thematic role assignment, which is central to sentence comprehension, requires processing the relation between nouns and verbs. In the SR clause, those relations can be processed sequentially; propositions can be processed one at a time. In contrast, in the OR clause segmentation is more difficult because the verbs, which are needed for assigning the roles, are concentrated at the end of the sentence. Consistent with this interpretation, the largest processing cost in OR sentences occurs at the moment that the verbs are displayed. This is the moment when noun-verb relations are extracted. Re-describing relative clauses in terms of relational complexity should allow researchers to entrench syntactic complexity into a cognitive literature that includes reasoning and problem solving (Halford & Andrews, 2004; Halford et al., 1998). For example, the maximum number of relations that can be processed simultaneously is four, according to estimates based experimental research.7 There is a developmental

7. The relational complexity of a task can be lowered by segmentation and by conceptual chunking, both of which are important aspects of expertise. Segmentation entails breaking tasks into less complex steps that can be processed serially. Conceptual chunking is the recoding of concepts into less complex relations. For example, ‘distance over time’ can be recoded as ‘speed’. Although useful in reducing complexity, conceptual chunking entails a loss of relational information.

 Diego Fernandez-Duque

progression to this maximum capacity (Andrews & Halford, 2002). It would be interesting to explore whether the developmental trajectory of relative complexity coincides with the developmental trajectory of relative clause use.

5.2 Matrix tasks When seen through the prism of relational complexity, matrix tasks share a resemblance to the transitive inference task just described. The best known example of this type of task is Raven’s Progressive Matrix (RPM), developed in 1938 as a measure of nonverbal intelligence. Over the years, the original version of the task has been adapted to accommodate different populations and methodologies (Carpenter, Just, & Shell, 1990). In a version of the task adapted for neuroimaging, a 3 x 3 matrix of figures is displayed with the bottom right figure missing (Christoff et al., 2001; Kroger et al., 2002). Participants have to infer the missing figure and select it from among a set of options. The complexity of the problem is based on the number of dimensions that relate figures to each other. For example, a 1-relational problem would vary in only one dimension (e.g., size). In one such problem, the figures might decline in size from left to right. A 2-relational problem would add a second dimension of change (e.g., shade). In this example, besides getting smaller from left to right the figures would also get darker from top to bottom. Solving a 1-relational problem requires evaluating only one dimension, while solving a 2-relational problem requires the simultaneous evaluation of two dimensions. In this sense, the RPM is conceptually similar to the transitive inference task, and to the processing of relative clauses. A comparison of 2-relational vs. 1-relational problems show activation of several brain areas, including anterior part of the left frontal cortex (area 10), posterior part of the IFG (area 44) and sub-cortical parts of the fronto-striatal loop (e.g., basal ganglia) (Christoff et al., 2001; Kroger et al., 2002).8

5.3 Dimensional change card sort task (DCCS) This task has been used successfully in the developmental literature to explore the development of executive functions (Zelazo, 2006). Children are asked to sort cards into two piles according to an explicitly stated rule, such as “Play the color game: if it’s red, it goes here [to the left], if it is blue it goes here [to the right]”. After several trials the rules change and children are explicitly told to sort based on a dimension

8. The IFG activation is related to the increased difficulty that is associated with more complex trials. In contrast, area 10 of the frontal lobe is specifically activated by relational complexity being active even in 2-relational trials in which the response was quick and accurate.

Cognitive and neural underpinnings of syntactic complexity 

that until then was irrelevant (e.g., ‘Now you will play the shape game: if it is a rabbit it goes here [left], if it is a boat it goes here [right]’). At the age of 3, most children are incapable of switching, even though they are able to verbalize the new rule if asked to do so. At the age of 5, most children are capable of performing the task. These findings have been interpreted in the context of a theory of cognitive complexity and control, which poses that rules are embedded in a hierarchical structure (P.D. Zelazo & Frye, 1998) . Application of the first-order rules (‘if red, left’, ‘if blue, right’; ‘if rabbit, left’, ‘if boat, right) is contingent on which higher order rule (‘sort by color’, ‘sort by shape’) is active based on the instructions. Although 3-year old children are capable of implementing two first-order rules, it is not until the age of five that they can coordinate them in an embedded structure. The embedded structure of the DCCS task raises the interesting question of whether it can be performed by people whose language, it has been argued, does not to include embedding (Everett, 2005). Said differently, the DCCS may be a good proxy for assessing the use of embedded structures and rules in speakers of that language. An advantage of the task is the simplicity of its instructions, which can be illustrated without words. For the purpose of this chapter, this task is relevant in that it requires the coordination of two different yet truthful statements about the very same object. Thus, the task requires flexibility in perspective taking. Something similar may be required for grasping that a noun phrase can be both the subject of the main clause verb and the object of the relative clause verb.

6. Conclusions The starting point of this chapter was the observation that OR clauses are often more difficult to process than SR clauses. In the literature, there are two main interpretations of this finding. One interpretation emphasizes the difference in syntactic operations between SR and OR clauses, arguing that the OR cost and the associated left IFG activation stem from the larger syntactic movement required by OR clauses (Grodzinsky, 2006). The other interpretation emphasizes syntactic complexity, arguing that the OR cost and IFG activation stem from increased demands in cognitive resources such as working memory. Although the bulk of the literature supports this latter interpretation, the exact nature of those resources remains a matter of dispute. Some researchers argue that the resources are for exclusive use in syntactic processing (Caplan & Waters, 1999) while others claim that the resources are more general, available also to nonsyntactic processing (King & Kutas, 1995). Finally, some researchers adopt a neutral stance, as if resources were a conceptual primitive in no need of further re-description (Gibson, 1998).

 Diego Fernandez-Duque

The literature reviewed in this chapter challenges the view that the resources for processing OR sentences are syntax-specific: overlapping brain regions are activated by complex tasks in both syntactic and non-syntactic domains. Furthermore, OR comprehension correlates with performance in non-syntactic verbal tasks. These tasks have in common their demands on verbal working memory, conflict resolution, and mental set switching. Further evidence that the resources for OR comprehension are available beyond syntax comes from neuroimaging studies in which the processing of relative clauses is compared to concatenated sentences (Booth et al., 1999; Just et al., 1996; King & Kutas, 1995). In those studies, the activation difference between OR and SR clauses has a very similar anatomy to the activation difference between SR and concatenated sentences. This suggests that the activation is not specific to a long syntactic movement across elements; otherwise, the SR clause and the concatenation sentences should have elicited the same amount of activation. The data also argue that the activation is not specific to the processing of embedded sentences; otherwise, OR and SR clauses should have elicited the same amount of activation, as both are embedded. The most parsimonious interpretation of this common anatomy of SR and OR clauses is, I would argue, that the increased activation in OR processing is driven by the increased difficulty — or the computational processes associated with this increase in difficulty. Of course, this interpretation is in line with the view that the resources are domain-general and closely related to executive functions. In further support of this view, executive function tasks with no syntactic demands often activate areas that overlap those activated by the syntactic task. These effortful tasks may require storing information in verbal working memory, resolving conflict among competitors, or switching one’s mindset, as reviewed in earlier sections of this chapter. In all such cases, left IFG is activated, as it is activated in the ‘OR vs. SR’ and the ‘SR vs. concatentation’ comparisons. The re-conceptualization of OR processing in terms of effortful executive functions helps address the underlying difference in the processing of OR and SR clauses. When difficulty between OR and SR clauses is equated, the left IFG activation disappears (Fiebach et al., 2004). This suggests that IFG activation is not triggered by the processing of OR per se, but rather by processes that are often — but not always — associated with those clauses. Arguing that the IFG activation is not due to OR processing per se but rather to the difficulty associated with those clauses may, at first glance, appear to be hair splitting. If OR clauses almost always are more difficult to process than SR clauses, then the two explanations for IFG activation would almost always be indistinguishable. This would be a compelling argument if OR clauses were consistently more difficult to process than SR clauses. On closer inspection, there are reasons to doubt this possibility.

Cognitive and neural underpinnings of syntactic complexity 

For example, when inanimate nouns are used in OR clauses, the OR cost is greatly reduced. This reduction is likely due to the frequent occurrence of those structures in natural language. For example, corpora analyses reveal that when the noun of the main clause is inanimate, the relative clause most likely to follow is an object-extracted clause (Roland, Dick, & Elman, 2007). Experimental evidence further supports this interpretation. When presented with a sentence that begins with an inanimate noun (e.g., the book that—) and asked to complete it, participants usually choose an objectextracted clause (Gennari & MacDonald, 2008).9 The reverse is true for animate nouns, in which case the sentence is more likely to include a SR clause. This modulatory role of animacy has often been ignored in studies of relative clause processing, which usually include only animate nouns. Ignoring the role of animacy has led to confounding clause type and frequency of use. The latter seems a likely contributor to difficulty of processing, as shown by studies in which executive functions are brought to bear in suppressing the dominant and most frequent interpretation. More generally, the re-conceptualization of ‘syntactic complexity’ in terms of executive functions (e.g., conflict resolution, set switching) helps to shed light on the otherwise mysterious concept of ‘resources’ to which many psycholinguistic studies have appealed when attempting to explain OR processing cost (Fernandez Duque & Johnson, 2002; Gibson, 1998). It also provides a common framework in which to entrench the study of syntactic complexity with studies of relational complexity and with cognitive neuroscience in general. As such, re-conceptualizing ‘syntactic complexity’ in terms of executive functions should prove useful to linguists and cognitive scientists alike.

References Amunts, K., Schleicher, A., Burgel, U., Mohlberg, H., Uylings, H.B. & Zilles, K. 1999. Broca’s region revisited: Cytoarchitecture and intersubject variability. The Journal of Comparative Neurology 412(2): 319–341. Andrews, G., Birney, D. & Halford, G.S. 2006. Relational processing and working memory capacity in comprehension of relative clause sentences. Memory and Cognition 34(6): 1325–1340. Andrews, G. & Halford, G.S. 2002. A cognitive complexity metric applied to cognitive development. Cognitive Psychology 45(2): 153–219.

9. This use of OR clauses is motivated by the speaker’s communicative intention to locate the referent of the inanimate noun in conversational space (Ford, Fox, & Thompson, 2003). Thus, OR clauses often have an “inanimate noun + that + pronoun” structure, with I or you as the pronoun. Of course, the referents of such pronouns have already been established in the discourse and as a consequence, these OR clauses are sometimes processed even more efficiently than SR clauses (Reali & Christiansen, 2007).

 Diego Fernandez-Duque Awh, E., Jonides, J., Smith, E.E., Schumacher, E.H., Koeppe, R.A. & Katz, S. 1996. Dissociation of storage and rehearsal in verbal working memory: Evidence from positron emission tomography. Psychological Science 7(1): 25–31. Baddeley, A. 1992. Working memory. Science New York, NY 255(5044): 556–559. Badre, D. & Wagner, A.D. 2007. Left ventrolateral prefrontal cortex and the cognitive control of memory. Neuropsychologia 45(13): 2883–2901. Ben-Shachar, M., Hendler, T., Kahn, I., Ben-Bashat, D. & Grodzinsky, Y. 2003. The neural reality of syntactic transformations: Evidence from functional magnetic resonance imaging. Psychological Science: A Journal of the American Psychological Society/APS 14(5): 433–440. Booth, J.R., Macwhinney, B., Thulborn, K.R., Sacco, K., Voyvodic, J. & Feldman, H.M. 1999. Functional organization of activation patterns in children: Whole brain fMRI imaging during three different cognitive tasks. Progress in Neuro-psychopharmacology & Biological Psychiatry 23(4): 669–682. Brass, M. & von Cramon, D.Y. 2002. The role of the frontal cortex in task preparation. Cerebral Cortex New York NY: 1991, 12(9): 908–914. Bush, G., Luu, P. & Posner, M.I. 2000. Cognitive and emotional influences in anterior cingulate cortex. Trends in Cognitive Sciences 4(6): 215–222. Caplan, D., Alpert, N. & Waters, G. 1998. Effects of syntactic structure and propositional number on patterns of regional cerebral blood flow. Journal of Cognitive Neuroscience 10(4): 541–552. Caplan, D., Alpert, N., Waters, G. & Olivieri, A. 2000. Activation of Broca’s area by syntactic processing under conditions of concurrent articulation. Human Brain Mapping 9(2): 65–71. Caplan, D., Chen, E. & Waters, G. in press. Syntactic and thematic effects on BOLD signal associated with comprehension and determination of plausibility of sentences with relative clauses. Caplan, D. & Waters, G.S. 1999. Verbal working memory and sentence comprehension. Behavioral and Brain Sciences 22(1): 77–126. Carpenter, P.A., Just, M.A. & Shell, P. 1990. What one intelligence test measures: A theoretical account of the processing in the Raven Progressive Matrices Test. Psychological Review 97(3): 404–431. Carpenter, P.A., Miyake, A. & Just, M.A. 1994. Working memory contraints in comprehension: Evidence from individual differences, aphasia, and aging. In Handbook of Psycholinguistics, M.A. Gernsbacher (Ed.), 1075–1122. San Diego CA: Academic Press. Chen, E., West, W.C., Waters, G. & Caplan, D. 2006. Determinants of bold signal correlates of processing object-extracted relative clauses. Cortex; A Journal Devoted to the Study of the Nervous System and Behavior 42(4): 591–604. Christoff, K., Prabhakaran, V., Dorfman, J., Zhao, Z., Kroger, J.K., Holyoak, K.J. et al. 2001. Rostrolateral prefrontal cortex involvement in relational integration during reasoning. NeuroImage 14(5): 1136–1149. Cooke, A., Grossman, M., DeVita, C., Gonzalez-Atavales, J., Moore, P., Chen, W. et al. 2006. Large-scale neural network for sentence processing. Brain and language 96(1): 14–36. Coulson, S., King, J. & Kutas, M. 1998. Expect the unexpected: Event-related brain response to morphosyntactic violations. Language and Cognitive Processes 131: 21–58. Crinion, J., Turner, R., Grogan, A., Hanakawa, T., Noppeney, U., Devlin, J.T. et al. 2006. Language control in the bilingual brain. Science New York NY 312(5779): 1537–1540. D’Esposito, M., Postle, B.R., Ballard, D. & Lease, J. 1999. Maintenance versus manipulation of information held in working memory: An event-related fMRI study. Brain and Cognition 41(1): 66–86.

Cognitive and neural underpinnings of syntactic complexity 

Derrfuss, J., Brass, M., Neumann, J. & von Cramon, D.Y. 2005. Involvement of the inferior frontal junction in cognitive control: Meta-analyses of switching and Stroop studies. Human Brain Mapping 2(51): 22–34. Derrfuss, J., Brass, M. & von Cramon, D.Y. 2004. Cognitive control in the posterior frontolateral cortex: Evidence from common activations in task coordination, interference control, and working memory. NeuroImage 23(2): 604–612. Dick, F., Bates, E., Wulfeck, B., Utman, J.A., Dronkers, N. & Gernsbacher, M.A. 2001. Language deficits, localization, and grammar: Evidence for a distributive model of language breakdown in aphasic patients and neurologically intact individuals. Psychological Review 10(84): 759–788. Dosenbach, N.U., Visscher, K.M., Palmer, E.D., Miezin, F.M., Wenger, K.K., Kang, H.C. et al. 2006. A core system for the implementation of task sets. Neuron 50(5): 799–812. Engle, R.W., Tuholski, S.W., Laughlin, J.E. & Conway, A.R.A. 1999. Working memory, shortterm memory, and general fluid intelligence: A latent-variable approach. Journal of Experimental Psychology: General 128(3): 309–331. Everett, D.L. 2005. Cultural constraints on grammar and cognition in Piraha. Current Anthropology 46(4): 621–646. Fan, J., Flombaum, J.I., McCandliss, B.D., Thomas, K.M. & Posner, M.I. 2003. Cognitive and brain consequences of conflict. NeuroImage 18(1): 42–57. Feredoes, E., Tononi, G. & Postle, B.R. 2006. Direct evidence for a prefrontal contribution to the control of proactive interference in verbal working memory. Proceedings of the National Academy of Sciences of the United States of America 103(51): 19530–19534. Fernandez-Duque, D. & Posner, M.I. 2001. Brain imaging of attentional networks in normal and pathological states. Journal of Clinical and Experimental Neuropsychology: Official Journal of the International Neuropsychological Society 23(1): 74–93. Fernandez Duque, D. & Johnson, M.L. 2002. Cause and effect theories of attention: The role of conceptual metaphors. Review of General Psychology 6(2): 153–165. Fiebach, C.J., Friederici, A.D., Muller, K. & von Cramon, D.Y. 2002. fMRI evidence for dual routes to the mental lexicon in visual word recognition. Journal of Cognitive Neuroscience 14(1): 11–23. Fiebach, C.J., Schlesewsky, M., Lohmann, G. von Cramon, D.Y. & Friederici, A.D. 2005. Revisiting the role of Broca’s area in sentence processing: Syntactic integration versus syntactic working memory. Human brain mapping 24(2): 79–91. Fiebach, C.J., Vos, S.H. & Friederici, A.D. 2004. Neural correlates of syntactic ambiguity in sentence comprehension for low and high span readers. Journal of Cognitive Neuroscience 16(9): 1562–1575. Fodor, J.A. 1983. The Modularity of Mind. Cambridge MA: The MIT Press. Ford, C.E., Fox, B.A. & Thompson, S.A. 2003. Social interaction and grammar. In The new psychology of language: Cognitive and Functional Approaches to Language Structure, Vol. 2, M. Tomasello (Ed.), 119–143. Mahwah NJ: Lawrence Erlbaum. Fox, B.A. & Thompson, S.A. 1990. A discourse explanation of the grammar or relative clauses in English conversation. Language 66: 297–316. Friederici, A.D. 2002. Towards a neural basis of auditory sentence processing. Trends in Cognitive Sciences 6(2): 78–84. Friederici, A.D., Fiebach, C.J., Schlesewsky, M., Bornkessel, I.D. & von Cramon, D.Y. 2006. Processing linguistic complexity and grammaticality in the left frontal cortex. Cerebral Cortex New York NY 1991 16(12): 1709–1717.

 Diego Fernandez-Duque Gennari, S.P. & MacDonald, M.C. 2008. Semantic indeterminacy in object relative clauses. Journal of Memory and Language 58(2): 161–187. Gibson, E. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition 68(1): 1–76. Gibson, E., Desmet, T., Grodner, D., Watson, D. & Ko, K. 2005. Reading relative clauses in English. Cognitive Linguistics 16(2): 313–353. Goel, V., Gold, B., Kapur, S. & Houle, S. 1997. The seats of reason? An imaging study of deductive and inductive reasoning. Neuroreport, 8(5): 1305–1310. Gordon, P.C., Hendrick, R. & Johnson, M. 2001. Memory interference during language processing. Journal of Experimental Psychology: Learning, Memory, and Cognition 27(6): 1411–1423. Gordon, P.C., Hendrick, R., Johnson, M. & Lee, Y. 2006. Similarity-based interference during language comprehension: Evidence from eye tracking during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32(6): 1304–1321. Grodzinsky, Y. 2006. A blueprint for a brain map of syntax. In Broca’s Region, Y. Grodzinsky & K. Amunts (Eds), 83–107. Oxford: OUP. Grodzinsky, Y. & Friederici, A.D. 2006. Neuroimaging of syntax and syntactic processing. Current Opinion in Neurobiology 16(2): 240–246. Grossman, M., Cooke, A., DeVita, C., Lee, C., Alsop, D., Detre, J. et al. 2003. Grammatical and resource components of sentence processing in Parkinson’s disease: An fMRI study. Neurology 60(5): 775–781. Grossman, M., Glosser, G., Kalmanson, J., Morris, J., Stern, M.B. & Hurtig, H.I. 2001. Dopamine supports sentence comprehension in Parkinson’s Disease. Journal of the Neurological Sciences, 184(2): 123–130. Grossman, M., Lee, C., Morris, J., Stern, M.B. & Hurtig, H.I. 2002. Assessing resource demands during sentence processing in Parkinson’s disease. Brain and language 80(3): 603–616. Grossman, M. & Moore, P. 2005. A longitudinal study of sentence comprehension difficulty in primary progressive aphasia. Journal of Neurology, Neurosurgery, and Psychiatry 76(5): 644–649. Hakes, D.T. 1972. Effects of reducing complement constructions on sentence comprehension. Journal of Verbal Learning and Verbal Behavior 11(3): 278–286. Hakes, D.T. & Foss, D.J. 1970. Decision processes during sentence comprehension: Effects of surface structure reconsidered. Perception and Psychophysics 8(6): 413–416. Halford, G.S. & Andrews, G. 2004. The development of deductive reasoning: How important is complexity? Thinking and Reasoning 10(2): 123–145. Halford, G.S., Wilson, W.H. & Phillips, S. 1998. Processing capacity defined by relational complexity: Implications for comparative, developmental, and cognitive psychology. The Behavioral and Brain Sciences 21(6): 803–831; discussion 831–864. Hamilton, A.C. & Martin, R.C. 2005. Dissociations among tasks involving inhibition: A singlecase study. Cognitive, Affective & Behavioral Neuroscience 5(1): 1–13. Hayes, A.E., Davidson, M.C., Keele, S.W. & Rafal, R.D. 1998. Toward a functional analysis of the basal ganglia. Journal of Cognitive Neuroscience 10(2): 178–198. Hazeltine, E., Poldrack, R. & Gabrieli, J.D. 2000. Neural activation during response competition. Journal of Cognitive Neuroscience 12 (Suppl 2): 118–129. Hochstadt, J., Nakano, H., Lieberman, P. & Friedman, J. 2006. The roles of sequencing and verbal working memory in sentence comprehension deficits in Parkinson’s disease. Brain and language 97(3): 243–257. Holmes, V.M. & O’Regan, J.K. 1981. Eye fixation patterns during the reading of relative-clause sentences. Journal of Verbal Learning and Verbal Behavior 20(4): 417–430.

Cognitive and neural underpinnings of syntactic complexity 

Just, M.A. & Carpenter, P.A. 1993. The intensity dimension of thought: Pupillometric indices of sentence processing. Canadian Journal of Experimental Psychology 47(2): 310–339. Just, M.A., Carpenter, P.A., Keller, T.A., Eddy, W.F. & Thulborn, K.R. 1996. Brain activation modulated by sentence comprehension. Science New York NY 274(5284): 114–116. Kaan, E. & Swaab, T.Y. 2002. The brain circuitry of syntactic comprehension. Trends in Cognitive Sciences 6(8): 350–356. Kan, I.P. & Thompson-Schill, S.L. 2004. Effect of name agreement on prefrontal activity during overt and covert picture naming. Cognitive, Affective & Behavioral Neuroscience 4(1): 43–57. Kane, M.J. & Engle, R.W. 2002. The role of prefrontal cortex in working-memory capacity, executive attention, and general fluid intelligence: An individual-differences perspective. Psychonomic Bulletin & Review 9(4): 637–671. Keller, T.A., Carpenter, P.A. & Just, M.A. 2001. The neural bases of sentence comprehension: A fMRI examination of syntactic and lexical processing. Cerebral Cortex New York NY: 1991, 11(3): 223–237. King, J.W. & Kutas, M. 1995. Who did what and when? Using word- and clause-level ERPs to monitor working memory usage in reading. Journal of Cognitive Neuroscience 7(3): 376–395. Kluender, R. & Kutas, M. 1993. Bridging the gap: Evidence from ERPs on the processing of unbounded dependencies. Journal of Cognitive Neuroscience 5(2): 196–214. Koelsch, S., Schulze, K., Sammler, D., Fritz, T., Muller, K. & Gruber, O. 2008. Functional architecture of verbal and tonal working memory: An FMRI study. Human Brain Mapping. Kroger, J.K., Sabb, F.W., Fales, C.L., Bookheimer, S.Y., Cohen, M.S. & Holyoak, K.J. 2002. Recruitment of anterior dorsolateral prefrontal cortex in human reasoning: A parametric study of relational complexity. Cerebral Cortex New York NY: 1991, 12(5): 477–485. Larkin, W. & Burns, D. 1977. Sentence comprehension and memory for embedded structure. Memory and Cognition 5(1): 17–22. Lewis, R. 1996. A theory of grammatical but unacceptable embeddings. Journal of Psycholinguistic Research 25: 93–116. Lewis, R.L., Vasishth, S. & Van Dyke, J.A. 2006. Computational principles of working memory in sentence comprehension. Trends in Cognitive Sciences 10(10): 447–454. Lindenberg, R., Fangerau, H. & Seitz, R.J. 2007. ‘Broca’s area’ as a collective term? Brain and Language 10(21): 22–29. MacWhinney, B. & Pleh, C. 1988. The processing of restrictive relative clauses in Hungarian. Cognition 29(2): 95–141. Mak, W.M., Vonk, W. & Schriefers, H. 2002. The influence of animacy on relative clause processing. Journal of Memory and Language 47: 50–68. Mayr, U., Diedrichsen, J., Ivry, R. & Keele, S.W. 2006. Dissociating task-set selection from taskset inhibition in the prefrontal cortex. Journal of Cognitive Neuroscience 18(1): 14–21. McNab, F., & Klingberg, T. 2008. Prefrontal cortex and basal ganglia control access to working memory. Nature Neuroscience 11(1): 103–107. Miller, G.A. & Chomsky, N. 1963. Finitary models of language users. In Handbook of Mathematical Psychology, Vol. 2, R.D. Lude, R.R. Bush & E. Galanter (Eds), New York NY: Wiley. Miyake, A., Carpenter, P.A. & Just, M.A. 1994. A capacity approach to syntactic comprehension disorders: Making normal adults perform like aphasic patients. Cognitive Neuropsychology 11(6): 671–717. Miyake, A., Friedman, N.P., Emerson, M.J., Witzki, A.H. & Howerter, A. 2000. The unity and diversity of executive functions and their contributions to complex ‘frontal lobe’ tasks: A latent variable analysis. Cognitive Psychology 41(1): 49–100.

 Diego Fernandez-Duque Monsell, S. 2003. Task switching. Trends in Cognitive Sciences 7(3): 134–140. Muller, H.M., King, J.W. & Kutas, M. 1997. Event-related potentials elicited by spoken relative clauses. Cognitive Brain Research 5(3): 193–203. Novick, J.M., Trueswell, J.C. & Thompson Schill, S.L. 2005. Cognitive control and parsing: Reexamining the role of Broca’s area in sentence comprehension. Cognitive, Affective and Behavioral Neuroscience 5(3): 263–281. Packard, M.G. & Knowlton, B.J. 2002. Learning and memory functions of the Basal Ganglia. Annual Review of Neuroscience 25: 563–593. Paulesu, E., Frith, C.D. & Frackowiak, R.S. 1993. The neural correlates of the verbal component of working memory. Nature 362(6418): 342–345. Peelle, J.E., McMillan, C., Moore, P., Grossman, M. & Wingfield, A. 2004. Dissociable patterns of brain activity during comprehension of rapid and syntactically complex speech: evidence from fMRI. Brain and language 91(3): 315–325. Poldrack, R.A., Wagner, A.D., Prull, M.W., Desmond, J.E., Glover, G.H. & Gabrieli, J.D. 1999. Functional specialization for semantic and phonological processing in the left inferior prefrontal cortex. NeuroImage 10(1): 15–35. Postle, B.R., Brush, L.N. & Nick, A.M. 2004. Prefrontal cortex and the mediation of proactive interference in working memory. Cognitive, Affective & Behavioral Neuroscience 4(4): 600–608. Reali, F. & Christiansen, M.H. 2007. Word chunk frequencies affect the processing of pronominal object-relative clauses. Quarterly Journal of Experimental Psychology 2006 60(2): 161–170. Roland, D., Dick, F. & Elman, J.L. 2007. Frequency of basic English grammatical structures: A corpus analysis. Journal of Memory and Language 57(3): 348–379. Thompson-Schill, S.L., D’Esposito, M., Aguirre, G.K. & Farah, M.J. 1997. Role of left inferior prefrontal cortex in retrieval of semantic knowledge: A reevaluation. Proceedings of the National Academy of Sciences of the United States of America 94(26): 14792–14797. Traxler, M.J., Williams, R.S., Blozis, S.A. & Morris, R.K. 2005. Working memory, animacy, and verb class in the processing of relative clauses. Journal of Memory and Language 53(2): 204–224. Van Dyke, J.A. & McElree, B. 2006. Retrieval interference in sentence comprehension. Journal of Memory and Language 55(2): 157–166. Wager, T.D., Jonides, J. & Smith, E.E. 2006. Individual differences in multiple types of shifting attention. Memory & Cognition 34(8): 1730–1743. Wager, T.D. & Smith, E.E. 2003. Neuroimaging studies of working memory: A meta-analysis. Cognitive, Affective & Behavioral Neuroscience 3(4): 255–274. Watson, D., Breen, M. & Gibson, E. 2006. The role of syntactic obligatoriness in the production of intonational boundaries. Journal of Experimental Psychology: Learning, Memory, and Cognition 32(5): 1045–1056. Watson, D. & Gibson, E. 2004. The relationship between intonational phrasing and syntactic structure in language production. Language and Cognitive Processes, 196, 713–755. Weckerly, J. & Kutas, M. 1999. An electrophysiological analysis of animacy effects in the processing of object relative sentences. Psychophysiology 36(5): 559–570. Yeung, N. Nystrom, L.E., Aronson, J.A. & Cohen, J.D. 2006. Between-task competition and cognitive control in task switching. The Journal of neuroscience : the official journal of the Society for Neuroscience, 265, 1429–1438. Zelazo, P.D. 2006. The dimensional change card sort DCCS: A method of assessing executive function in children. Nature Protocols 1(1): 297–301. Zelazo, P.D. & Frye, D. 1998. Cognitive complexity and control: II. The development of executive function in childhood. Current Directions in Psychological Science 7(4): 121–126.

Neural mechanisms of recursive processing in cognitive and linguistic complexity Don M. Tucker1,2, Phan Luu1 & Catherine Poulsen1 1Electrical

Geodesics, Inc/2Psychology Department and NeuroInformatics Center, University of Oregon Cognition in the human brain requires processes of memory consolidation and retrieval that are carried out across reentrant connections between limbic cortex and multiple network levels of the neocortex. Given this layered architecture, and the point-to-point reentrance of the connections, cognition is likely to be recursive, changing its internal representations dynamically with each cycle of consolidation. To provide structure and constancy within this dynamic interplay, language operations appear to draw on the capacity for inhibitory specification emergent within the ventral, paleocortical corticolimbic pathways. We propose that inhibitory specification has been essential to regulate the dynamism of recursive consolidation, supporting the evolution of both the object qualities of words and the regularized structure of grammar.

1. Introduction Complexity in grammatical clause organization provides a hierarchic organization that allows efficient constructions of meaning. A provocative proposal in recent years has been that the inclusion of recursive structures, in which a clause is included within a similar type of clause, allowed the emergence of modern language (Hauser, Chomsky, & Fitch, 2002). Clause nesting is certainly important in complex language constructions. However, in the conventional terms of information processing, a nested structure does not require recursion, unless processing actually operates within the structure as well as outside it, resulting in a change of the representation at each structural level. The difference is significant: supporting true recursion is a challenge for computer languages, whereas nesting is a straightforward extension of even simple computer languages and compilers. Recursion is almost certainly challenging for brains as well as computers. Yet it appears to be integral to the information processing of multi-level reentrant neural networks that we find in the mammalian brain. In this paper, we consider how complexity may arise in corticolimbic representations, through recursive interactions between linked networks. In sensory systems, recursive interactions are known to be required for memory consolidation (Squire, 1986; 1998), linking the feature extraction

 Don M. Tucker, Phan Luu & Catherine Poulsen

in sensory networks to the motive and memory representations in limbic networks (Tucker & Luu, 2006). We consider how the recursive processing in memory may be required in organizing language, both in the general task of pragmatic comprehension and in the specific task of generating syntactic complexity. Complexity in memory organization is essential to forming language representations, and to making use of the very limited capacity of human working memory. That human memory capacity is limited is a fact that is often experienced in daily life, yet it may not be fully apparent from introspection alone. In section 2, we review both classical and recent experimental psychology evidence documenting the limits of human memory. This evidence provides an important basis for understanding not only the cognitive constraints that must be addressed by linguistic constructions, but how the emergence of these linguistic constructions allowed humans an efficient organization of memory elements and thus novel reasoning abilities. Recent findings and theoretical models in neuropsychology have suggested that memory is achieved through dual neural systems, each of which provides unique representational properties, but also unique limitations. In section 3 we review the evidence on the unique dorsal and ventral corticolimbic pathways for consolidating memory through recursive, reentrant network interactions. The dorsal corticolimbic circuit centered on the hippocampus and cingulate gyrus supports configural memory. The ventral corticolimbic circuit centered on the amygdala and anterior temporal, insular, and orbital frontal regions supports item or object memory. Traditional evidence on aphasia syndromes emphasizes the importance of object memory to both expression and comprehension of language. The ventral limbic pathway’s unique capabilities for specifying objects, with unique features inhibitory control, may be integral to the left hemisphere’s capacity for specifying denotative semantics generally, and for creating complex linguistic constructions with the aid of grammatical conventions. Nonetheless, meaningful grammatical constructions may depend on both corticolimbic memory systems, with each one contributing unique abilities in representation and control. Within the potentially chaotic dynamics of recursive, reentrant network interactions, the ability to organize a complex linguistic structure such as a hierarchic clause may depend upon the capacity for what may be called inhibitory specification, in which certain meanings are isolated within working memory, such that they can be sustained and grouped within hierarchic structures. Clues to the neural mechanisms of inhibitory specification can be gained from studying the sequencing and routinization of action within the motor system, and from examining the unique properties of object memory wtihin the ventral corticolimbic pathway. Nonetheless, complexity in language requires that the specified semantics and syntax of the ventral corticolimbic networks is framed within the pragmatic representations of the context of experience and communication formed within the dorsal

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

corticolimbic pathways. In section 4 we consider the differing forms of motor control within dorsal and ventral pathways that suggest unique cybernetic biases are integral to the differing forms of memory consolidation. In the concept of action regulation we consider the fusion of motive controls, motor cybernetics, and cognitive representations that are formed through recursive interactions between limbic and neocortical networks in the ongoing process of memory consolidation. These generic mammalian mechanisms of learning, memory, and behavior are the foundations for language, and language complexity. In section 5 we consider how the classical and modern studies of aphasia reveal the dependence of language on multiple levels of representation. These include the dynamic interplay between expressive and receptive functions, but also the interplay between the visceral, motive control of consolidation in limbic networks and the somatic articulation of perceptions and actions in primary sensory and motor cortices. In between these visceral and somatic boundaries of consolidation the unimodal and heteromodal association cortices provide integrative way stations for intermediate levels of recursive transformation of language representations. The result of the dialectical interplay across these four reentrant network levels (limbic, heteromodal association, unimodal association, and primary sensory or motor) may be an arbitration of visceral and somatic demands that allows abstract cognitive and linguistic representations. In section 6 we suggest that complex grammatical structures in language are a key mechanism for abstract thought, in which the requirements for somatic articulation of action within the motor system arbitrate with the internal, visceral motivational control of meaning in the process of communication.

2. Limited capacities of representation and binding Cognitive-experimental research on memory limitations reaches back more than 50 years, to early studies marked most notably by Miller’s now famous estimation of short-term memory (STM) capacity at 7 ± 2 chunks of information (Miller, 1956). Miller also observed that recoding of information into ever larger “chunks” is instrumental in expanding the capacity of STM, emerges naturally with experience, learning and expertise, and is ubiquitous in language. At lower levels of language processing, for example, chunking is evident in the recoding of phoneme sequences into syllables, syllables into words, and words into phrases. At a higher level, sequences of coordinated phrases can be organized into hierarchic-subordinate structures that enable the more efficient expression and comprehension of increasingly complex ideas encoded into linear discourse. By the early 70s, the construct of STM was further refined by the introduction of working memory (WM) models (Baddeley & Hitch, 1974). Like STM, WM is characterized

 Don M. Tucker, Phan Luu & Catherine Poulsen

by capacity limitations, but here, they are distributed across both on-line storage and processing functions. One important corollary of this view is that the effective capacity of WM can be increased by efficiency of storage and/or processing components. The distinction between controlled and automatic processing (Schneider & Chein, 2003; Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977) is central here. Controlled processes tend to be slow and effortful, require attentional resources, and are deployed intentionally early in learning or in novel contexts. In contrast, automatic processes are fast, require minimal attentional resources, and are carried out in a ballistic fashion following extensive experience. Just as recoding and chunking can increase the amount of information maintained in WM, a shift to automatic processing can increase the pool of available cognitive resources. As experience in a language accrues, lower-level processing components such as word recognition, lexical access, or routine syntactic parsing become automatized, freeing up resources to be dedicated to higher level processes, such as integrating information across phrase boundaries and processing more complex relations among arguments. A second implication of WM as both storage and processing functions is its relevance to goal-directed action. One typically doesn’t just hold information in STM for later recall; one does something with that information. In other words, information is selectively maintained and manipulated in working memory in order to enhance adaptive behavior. From this perspective, WM is closely aligned with motivation and self-regulated action. An influential capacity view of attentional resources, the selection-for-action theory (Allport, 1985), posits that processing limits occur not at the level of perception and sensory selection, but out of the need to engage in coherent (usually sequential), behavioral responses — either as covert action, or internal thought. This can lead to selective, top-down enhancement of action-relevant sensory attributes (Hannus et al., 2005) and, conceivably, configural information, in order to bias their active maintenance in WM, and suggests that the contents of WM are not driven primarily by bottom-up sensory selection. Working memory thus may play an important role in the integration or binding of sensory information with action regulation. Empirical research on the impact of memory limitations on syntactic complexity has been most extensively studied by relating individual differences in WM capacity with the comprehension of syntactically complex sentences. The reading-span task (Daneman & Carpenter, 1980) has been frequently used to assess an individual’s combined WM storage and processing capacity in the language domain. King and Just (1991) demonstrated that individuals with relatively low WM capacity for language (as measured by the reading-span task) exhibited longer reading times and poorer accuracy than higher-capacity readers for complex sentences with an object-relative clause (e.g., The reporter that the senator attacked admitted the error), but they performed comparably on simpler sentences with a subject-relative clause (e.g., The reporter that attacked the senator admitted the error). Here, the more

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

processing-intensive object-relative clause requires the reader to associate the head noun with two syntactic roles (e.g., “reporter” as subject of the main clause and object of the relative clause), whereas the subject-relative clause requires activation of only one role (that of subject) for both the main and relative clauses. A more recent fMRI study (Prat, Keller, & Just, 2007) similarly found that lowcapacity readers had slower reading times than high-capacity readers for both activeconjoined (The writer attacked the king and admitted the mistake at the meeting.) and object-relative sentences (The writer that the king attacked admitted the mistake at the meeting.). Accuracy, however, was significantly worse in low-capacity readers only for the object-relative sentences. Thus, assuming that WM capacity is the major factor in individual differences in reading capacity, slower reading times appeared able to compensate for low WM capacity when reading syntactically simple sentences, but comprehension remained impaired on syntactically complex sentences. Functional MRI results suggested that processing was both less efficient and poorly coordinated in low-capacity readers. Specifically, low-capacity readers had higher BOLD responses than high-capacity readers (particularly in frontal control, and occipital regions) suggesting they consumed more resources despite poorer performance. Functional connectivity analyses further indicated poorer synchronization among left-hemisphere language regions, including Broca’s and Wernicke’s areas, in low-capacity readers. The neuroimaging data may thus provide insight into mechanisms for the concept of WM, showing that low-capacity readers engaged more activity in, but less coordination among, WM-related brain areas. In light of these and similar findings, Just and Varma (2007) proposed that working memory capacity limits may be understood in terms of resource constraints on neural activation. That is, the effective limits of working memory may reflect the capacity to recruit multiple neural regions that sustain activation of context-relevant computations or information, while at the same time coordinating communication across this dynamic network in the service of self-regulated action or thought. Although full explication of this approach is beyond the scope of this chapter, of relevance here is their 4CAPS model of sentence comprehension. Although the authors acknowledge the contribution of other brain regions, their model focuses on the respective roles of Wernicke’s and Broca’s areas. They propose that Wernicke’s area, the “associative center,” specializes in retrieving relevant, language-based associations, prior knowledge (procedural and declarative) and perceptual inputs and in configuring “designs” or templates for new language representation. Broca’s area, the “structure-builder” center, then takes these loosely structured templates and their associated information (in essence the information actively maintained in WM), and builds them into recognized syntactic structures. Finally, it may be important to keep in mind that memory capacity limits not only the number of representational elements (words, clauses) but the number of interrelations (or bindings) among those items that can be kept active in WM

 Don M. Tucker, Phan Luu & Catherine Poulsen

(Halford, Baker, McCredden, & Bain, 2005; Halford, Cowan, & Andrews, 2007). Because memory is required to maintain binding among the elements, Halford’s reasoning suggests that the effective number of items active in a linguistic frame may be reduced from the “magical number” seven, down to only 3.5, on average. Binding requirements may be particularly relevant when considering capacity limitations for hierarchical embedding. The essential cognitive work to be done in comprehending such structures is to identify the relevant constituents and establish their interrelationships (i.e., who did what to whom). In this context, it is interesting to note that a recent corpus linguistic analysis across seven European languages (Karlsson, 2007) concluded that the maximum number of center-embeddings employed is three. This is in close agreement with Halford’s estimate of the number of elements and their relations that can be maintained in WM. In summary, we propose that the relationship between working memory and syntactic processing may be associated with the need to translate a holistic relational structure — which is formulated at a global, syncretic level — into extended, linear discourse. The linear nature of linguistic expression entails that the constituent elements of this relational structure be actively maintained in working memory for integration to occur over a time delay. This may require both sustained activation of relevant processing regions and coordination of processing across those regions. Hierarchic-subordinated syntactic structures offer more efficient and elaborated articulation of such complex relations than do sequential-coordinated structures, but they increase memory demands for on-line retention and integration of multiple referents. Working memory capacity may, therefore, both enable and constrain syntactic complexity. While the ability to hold multiple referents in working memory facilitates the production and interpretation of complex sentences, inherent capacity limitations of working memory appear to place constraints on the level of complexity (e.g., number of referents, distance to resolution) that can be generated and interpreted successfully (Gibson, 1998).

3. Corticolimbic recursion in memory consolidation The limits on memory representation, and on the binding of information elements, may be better understood by examining the dual corticolimbic control systems that guide memory consolidation. Modern neuroanatomical studies have shown that the neocortex evolved from the limbic cortex in a nested structure (Pandya & Seltzer, 1982; Pandya & Barnes, 1987; Pandya & Yeterian, 1984) in which each of four levels of neocortical differentiation emerged embedded within its predecessor (Figure 1). Within limbic (or paralimbic) cortex emerged the heteromodal “association” cortex, then the unimodal association cortex, and finally the modality specific sensory or

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

motor cortex (Mesulam, 2000). One primary pattern of connectivity is between levels, with each level connecting to its adjacent, embedded or embedding, neighbor with reentrant bidirectional projections (Pandya & Seltzer, 1982). The result of these several levels of interconnectivity is to create a “pathway” such that visual information, for example, is processed from primary visual areas (which receive thalamic projections) to secondary visual association cortex, to heteromodal association cortex, then to limbic cortex (Figure 1, right). In the frontal lobe, the reverse direction of network embeddedness obtains, such that actions are initially organized on a limbic base (in orbital frontal or anterior cingulate networks), and are progressively articulated through frontal heteromodal association, then premotor association, and finally primary motor cortices (Figure 1, left).

Figure 1. Dorsal (upper) and ventral (lower) pathways of the human brain, showing primary sensory/motor (diagonal lines), unimodal (dashed), heteromodal (stippled), and limic (shaded) cortices. At left the arrows show the origin of motor control in dorsal and ventral limbic regions, with progressive organization of actions through the four linked networks in each pathway. At right the arrows show the initial processing of visual data in primary visual cortex, and the elaboration within memory through processing in the four network levels of the dorsal and ventral visual pathways. © 2007 Don M. Tucker.

Although the arrows in Figure 1 show the typical processing assumption with vision in and motor out, this fails to capture the recursive nature of the actual neural processing. For both sensation and action, the connections are reentrant in that processing is not just one-directional. For vision, for example, there are as many

 Don M. Tucker, Phan Luu & Catherine Poulsen

connections proceeding from limbic cortex out to heteromodal, to unimodal, and finally to primary visual cortex as proceed in the opposite direction. Reentrance is particularly important in building cognitive or linguistic models, because it emphasizes the distributed nature of representations that are recreated in the multiple instantiations across the embedded networks. Reentrance describes the structural connectivity, and it implies the processing, recursion, that is implicit with this point-to-point connectivity. Information processing must proceed in some form, in some unknown recursive fashion, in both directions in each sensory or motor corticolimbic pathway (Tucker & Luu, 2006). Figure 2 shows a medial view of the right hemisphere, with a schematic illustration of the global pattern of connections within the hemisphere. The four network levels shown in Figure 1 are shown in the middle panel of Figure 2 as they would be seen in the folded cortex, and in the bottom panel as they might be seen if each dorsal and ventral sensory and motor corticolimbic pathway were unfolded. As shown by the quantitative anatomical studies of Pandya and associates (Pandya & Seltzer, B, 1982; Pandya & Barnes, 1987; Pandya & Yeterian, 1984), the greatest connection density is between adjacent networks within a pathway (e.g., unimodal dorsal visual to heteromodal dorsal visual), with network-to-network reentrance of these connections implying that processing must be recursive in some way. The lines between pathways illustrate the principle that cortical connections tend to stay between levels (e.g., hetero modal to heteromodal), and that the density of interregional connections is greatest at the limbic level and virtually absent (except for motor-somatosensory connections of the body map) at the primary sensory and motor neocortices (which connect primarily to subcortical and the adjacent unimodal cortex). Figure 3 illustrates the remarkable organized topography of the withinpathway connections (after Pandya & Yeterian, 1984). These are the connections that mediate the process of memory formation. In vision, for example, this pattern of connections allows representations of motive significance in limbic networks to consolidate the perceptual data over time to organize visual experience. The corticolimbic connections are reentrant, with outputs to core hemispheric (heteromodal, limbic) networks feeding back to the more lateral neocortical (unimodal, primary sensory) networks, with an ordered topography of this reentrance suggesting a specific structure to the recursive processing. The neocortical => limbic direction of (“forward”) projections arise within the superficial layers, which are thought to organize local processing. The limbic => neocortical direction of (“back”) projections arise within infragranular layers, which regulate subcortical communications, including to the thalamus. Although we do not know its neurophysiological nature, we do know that this reentrant and recursive corticolimbic processing is required for memory consolidation. Primate and rodent studies have shown that sensory data must be processed across all

Neural mechanisms of recursive processing in cognitive and linguistic complexity  Cerebral cortex (telencephalon) Thalamus and hypothalamus (diencephalon) Midbrain (mesencephalon) Pons and cerebellum (metencephalon) Medulla (myencephalon)

Figure 2. Schematic of the connectional architecture of a cerebral hemisphere. © 2007 Don M. Tucker

 Don M. Tucker, Phan Luu & Catherine Poulsen

Figure 3. Schematic of within-pathway corticolimbic connectivity (after Pandya). Top: view of a pathway, e.g., ventral visual pathway. Middle: cortical => limbic projections, with primary connections proceeding from supragranular layers toward granular layers, as if each more diferentiated network level evolved to model and replace the thalamic input. Bottom: the limbic => cortical “back” projections are as dense as the “forward” projections, and quantitative studies show the majority proceed from infragranular toward supragranular layers of the cortex. © 2007 Don M. Tucker

levels of the corticolimbic pathway to be consolidated in memory (Squire, 1986, 1998); disconnection of sensory and association cortex from the limbic base results in a profound deficit in consolidating new learning. Importantly, even though connectivity is broken across the corticolimbic pathway, prior memories may be accessed to guide behavior, depending on the intact connectivity of the residual networks. Although not as anatomically explicit as the animal evidence, the evidence on human amnesia is consistent with this general outline, such that specific agnosias are observed with lesions to association cortex in a given modality, and general amnesia is observed with lesions of limbic networks of the medial temporal lobe (Squire, 1986, 1998).

3.1 Thalamic resonance of the efferent copy An analysis of the nested corticolimbic networks thus provides an interesting theoretical basis for understand the levels of representation in human language (Luu & Tucker,

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

1998; Tucker, Frishkoff, & Luu, 2008). However, it has long been apparent that any theory of cognitive and linguistic representation based on anatomical connectivity must consider the extensive network created by thalamocortical, and corticothalamic, projections (Crosson, 1999). Recently, research into the anatomy and function of thalamic connections has suggested that most if not all of thalamic afferents (input connections) reflect copies of motor control projections to subcortical circuits (Guillery & Sherman, 2002; Sherman & Guillery, 2002). As a result, thalamocortical projections would then reflect processes of action monitoring. With extensive intrathalamic connections providing modulatory control over this bidirectional traffic, cortical control over the thalamus can be understood as a key mechanism for attentional control of behavior, and control of the sensory representations that guide behavior (Guillery & Sherman, 2002; Sherman & Guillery, 2002).

3.2 Limbic-diencephalic learning circuits In the mechanisms underlying the language process, both corticolimbic and corticothalamic networks must be integrated in some fashion to allow executive control of working memory. This integration must allow the elements of communication (agents, acts, objects) to be not only maintained in the minds of the speaker and listener but also bound in meaningful linguistic patterns. One insight to this integration comes from evidence that memory is achieved by two different cortico-limbic-thalamic circuits, each with a unique learning strategy. These learning strategies can be seen as different ways of optimizing the use of limited memory capacity. Each of these circuits supports a different set of the nested corticolimbic networks. The circuit centered on the hippocampus supports the dorsal corticolimbic pathway, with its primary association cortices in the parietal regions of the posterior brain and mediodorsal regions of the frontal lobe. The circuit centered on the amygdala, pyriform cortex, and insula supports the ventral corticolimbic pathway, with its primary association cortices in the occiptotemporal regions of the posterior brain and ventrolateral regions of the frontal lobe. It is the ventral pathway that appears particularly important to inhibitory specification of meaning in language, and thus to applying the constraints on neural network recursion that allow syntactic complexity.

3.2.1 Configural representations and context-updating Papez (1937) described a circuit engaging the hippocampus, posterior cingulate cortex, ventromedial thalamus, and mammalary bodies of the hypothalamus that readily sustained seizures and appeared to be important to the motivational control of behavior. Modern memory research has shown this circuit, supporting the dorsal corticolimbic pathway, to be critical to spatial memory in rodents, and very likely to configural representations in humans (Nadel, 1991).

 Don M. Tucker, Phan Luu & Catherine Poulsen

In addition to being specialized for a holistic representation of the spatial or configural context, the dorsal circuit appears to be specialized to control learning in a unique way. Animal studies (Gabriel et al., 1983; Gabriel, Sparenborg, & Kubota, 1989) suggest that the dorsal circuit adapts gradually and more or less passively to changes in the environmental context, a process that can be described as contextupdating (Luu, Flaisch, & Tucker, 2000; Luu & Tucker, 2003). This can be seen as a primitive form of associative learning, but one that is well suited to maintenance of a holistic internal model of the current perceptual and behavioral context.

3.2.2 Object representations and sustained focus In contrast, a second cortico-limbic-diencephalic circuit centered on the amygdala and ventrolimbic networks engages the mediodorsal thalamus and supports the item or object memory representations of the ventral pathway (Aggleton & Brown, 1999; Mishkin, 1982). Supplying input to the orbital frontal and ventral (subgenual) anterior cingulate cortex, the ventral limbic regions provide not only consolidation of object perceptions, but organization of motor control directed by this consolidation. The specific control properties of this circuit and associated networks are suggested by animal learning studies by Gabriel and his associates. These studies found that lesions of the anterior cingulate cortex impair the animal’s ability to adapt rapidly to changing circumstances (Gabriel et al., 1983; Gabriel et al., 1989). The ventral pathway seems uniquely able to detect conflict or incongruity with the current context model, and then maintain a focus of attention to organize new actions required by the discrepancy (Tucker & Luu, 2007).

3.3 A limbic base for consolidation Why are these differing learning strategies manifested by different corticolimbic circuits? One answer may be functional or algorithmic, explaining the adaptive advantages of different control biases. Studies of robotic control have shown that certain cybernetic (representation and control) designs allow efficient internal guidance of action, in a feedforward fashion (Hendler, 1995). However, these designs respond poorly to unanticipated changes in the environment. Other designs incorporating feedback control are more responsive to changing action plans when events intercede, but they are poor at maintaining goal-directed actions. Because these alternate cybernetics may require fundamentally different neural mechanisms, mammalian evolution seems to have instantiated them in different learning circuitry. Another answer is neurophysiological, and it comes from an analysis of the differing autonomic or bodily self-control functions carried out by the dorsal and ventral limbic networks. Neafsy and associates (Neafsey, Terreberry, Hurley, Ruit, & Frysztak, 1993) have shown that the cingulate cortex at the base of the dorsal corticolimbic pathway

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

carries out visceromotor regulation of internal bodily functions and associated motivated behavior. Because visceromotor regulation emerges directly and reflexively from the hypothalamic and limbic monitoring of bodily states, this form of control may be the primitive basis for the feedforward learning strategy that appears integral to the more extended consolidation operations of the dorsal corticolimbic pathway. In contrast, the insular cortex and associated ventral limbic networks appear specialized for viscerosensory regulation of internal functions and associated motivated behavior (Neafsey et al., 1993). This basis in visceral control may be consistent with the feedback control mode integral to the ventral pathway’s ability to detect discrepant events and to maintain focused attention to redirect adaptive actions.

3.4 The visceral basis of pragmatic and semantic memory systems In examining the connectional architecture of limbic networks, we find them densely interconnected across modalities, compared to the isolated modules of somatic representation in sensory and motor cortices (D.N. Pandya & Seltzer, 1982). This suggests that limbic networks must provide the brain’s most integrative representations, in contrast to the traditional view that these integrative representations are formed in “association” cortices such as lateral frontal, temporal, and parietal regions. Jason Brown (1977, 1988) pointed to clinical studies of aphasia that appear consistent with this connectional evidence, indicating that whereas lesions of neocortex (including Broca’s and Wernicke’s areas of heteromodal association cortex) would produce comprehension and expression deficits, it is only with lesions of limbic cortex that patients suffer severe semantic deficits. At the same time as providing a holistic level of representation, limbic cortex (dorsal division based in parahippocampal and cingulate; ventral division based in periamygdalar, anterior temporal, and insular) is responsible for visceral, homeostatic and motivational, functions (Neafsey et al., 1993). The implication may be that memory consolidation within these temporal-limbic networks is both highly integrative and motivationally significant. The functional specializations of the dorsal and ventral corticolimbic pathways may imply coordinated and unique contributions to language (Jeannerod, 1995). The key features of linguistic representation, with objects in the form of words capturing meaning, then structured in predictable sequences, draw on the cybernetics of the ventral corticolimbic pathway, both in the comprehension networks linked to the ventral temporal lobe, and the expression networks centered in the inferior frontal lobe. The culturally ritualized conventions of language objects represented in ventral networks must provide the basis for semantics. On the other hand, the configural information skills of the dorsal corticolimbic networks may be more important to pragmatic representational functions, through linking the specific

 Don M. Tucker, Phan Luu & Catherine Poulsen

language units and structures to the shared understanding of the social and environmental context of the communication.

4. Motive and complexity in representational actions Although perhaps integrative, the diffusely interconnected representations of limbic networks are likely to be poorly differentiated. Cognition, and linguistic structure, formed at this level can be seen as syncretic, with multiple referential implications fused within a primitive connotative binding. To understand how more differentiated linguistic patterns, including the more traditional denotative semantics, can emerge from this elemental connectional matrix, it may be helpful to consider how actions are organized to mediate between bodily needs and environmental constraints. This mediation between limbic and neocortex is critical to memory consolidation, and it involves recursion between multiple adjacent cortical networks (limbic, heteromodal, unimodal, primary sensory or motor). We propose that by understanding the recursive mechanisms of organizing memory capacity, specifically in the context of action planning, we can gain insight into the neurocognitive process that generates, and that benefits from, grammatical complexity.

4.1 Complex constructions in the sensorimotor machine One of the first to recognize the psychological significance of the evolutionarydevelopmental order of the brain’s anatomy, Hughlings Jackson emphasized that all the brain’s functional circuits are linked to sensory or motor operations, such that the brain can be seen as a “great sensorimotor machine” (Jackson, 1931). At first glance, this pithy summary appears to be too simple to help students of the brain understand the complexity of its functions, including language representation and organization. However, we suggest that Hughlings Jackson’s summary can be taken as a directional pointer, emphasizing that cognition does not arise ex cathedra from the vapors, but rather emerges from the body’s sensorimotor, and visceral, operations. From this perspective, basic mechanisms of action regulation suggest insight into how cognitive expressions of the mind arise from more elementary neural processes.

4.2 Chunking of action sequences and attentional capacity Actions must be sequenced to organize coherent behavior. In his analysis of the “problem of serial order” Lashley explained that the associative chaining of traditional behaviorism could not account for even elementary challenges handled by the mammalian motor system (Lashley, 1951). As a result, a representational theory is required. To understand the hierarchic structure within embedded clauses in language, it may be useful to begin

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

with a basic analysis of how actions are grouped within clusters or chunks, which are then able to be executed with minimal demands on attentional capacity (Keele 1981; Keele & Hawkins, 1982). A similar efficiency appears to be provided by complexity in language structure, allowing not only the speaker but the listener to automatically capture nested semantic packets, thereby freeing attentional capacity for broader interpretation of the discourse and its context. A neuropsychological analysis of action planning shows there are dual routes to sequencing actions, one in dorsal cortex and one in ventral cortex, each emerging from a diffferent basis in the limbic system to shape action plans of the frontal lobe, and each providing a unique form of action regulation. By analyzing the limbic circuits that give rise to these cortical systems, it is possible to frame each mode of action regulation within a more general cognitive framework describing dual modes of memory consolidation. Through extending this neuropsychological analysis, we will argue that complexity in language relies on dual modes of motor control that are fundamental to organizing cognitive and linguistic structure generally.

4.3 The visceral basis of the motive-memory Between the visceral representations of needs and motives and the somatic representations of sensory inputs and motor outputs are processes of inter-network recursion that give rise to what psychologists would consider higher-level cognition, including such constructs as executive control and working memory. How can we understand these processes within an action-regulation framework? Yakovlev (Yakovlev, 1948) provided a key insight when he proposed that the evolution of language can be seen as another extension of the more general evolution of motility. In primitive brains, such as that of salamanders, movement and homeostatic control are closely linked within brainstem structures, such as the tectum and tegmentum (Herrick, 1948). Actions are characterized by core, axial movements, reflecting the holistic (and undifferentiated) nature of actions and motivation. They emerge directly from internal motive processes to engage the external enviornmental process. For Yakovlev, the general organization of motility is a process of “exteriorization,” as the internal urge is manifest in actions contacting the world. Yakovlev viewed language within this general framework of motility, thereby providing a theoretical model for understanding communication as bounded both by biological needs and the constraints of action regulation. This organization of actions and motivation within the primitive brains of amphibians stands in contrast of course to that of mammalian brains, wherein motor control spans a more complicated hierarchy that includes the neocortex. Yet, with increasing complexity in brain organization, the translation of motivational influence to action remains central to adaptive behavior. A particularly illuminating example is the seemingly inappropriate behaviors of monkeys with lesions to the amygdala, producing the Kluver-Bucy

 Don M. Tucker, Phan Luu & Catherine Poulsen

syndrome. These monkeys approach all objects without fear and react to them as if they are novel, and they also demonstrate inappropriate behaviors to peers. Pribram (Pribram, 1991) noted that this syndrome reflects the lack of visceral familiarity that usually imbues sensory experiences; without intact visceral-sensory associations, actions become not only amnestic but dysregulated. Pribram refers to the contribution of diencephalic and limbic structures as a protocritic function — holistic, undifferentiated, and motivationally relevant — that gives meaning to actions. Pribram’s theory suggests how the viscerosensory functions of the insula, amygadala, and associated ventral limbic networks (Neafsy, et al., 1993) may be integrated within the motive basis of action regulation. Complementing the viscerosensory function of the ventral limbic trend is the visceromotor control integral to the dorsal limbic circuit (Neasfy et al., 1993). The unique symptoms of lesions to the dorsal limbic core of the hemisphere have long been known (Barris & Schuman, 1953). Bilateral anterior cingulate lesions result in akinetic mutism, a condition in which the patient exhibits little to no spontaneous action, and yet is able to react with coherent action and cognition when prompted. We can infer from this deficit that the dorsal limbic base of the frontal lobe is normally involved in the spontaneous, motivated direction of actions to the world.

4.4 Projectional and feedback modes of action regulation From these dual foundations in motivated operations of memory consolidation, actions must emerge. Modern functional and anatomical analyses have suggested that these viscerosensory and visceromotor divisions of the mammalian cortex evolved from the primitive telencephalon of birds and reptiles, each division applying different control properties in the organization of action. In addition to supporting configural cognition and memory for the spatial context of behavior (Mishkin, 1982; Aggelton & Brown, 1999), the dorsal pathway supports a projectional, feedforward mode of action, in which behavior is launched toward a goal (Goldberg, 1985) (Passingham, 1987; Shima & Tanji, 1998). Animal studies show that cells within the SMA are preferentially active (and fire several seconds before the actual movement) when actions are initiated from memory, as opposed to when they are guided by sensory input (Mushiake et al., 1990). In humans, patients with SMA lesions can reproduce sequences of actions when they are visually cued but can not produce the same actions from memory (Halsband et al., 1993). These observations are consistent with the notion of projectional, ballistic control in the dorsal pathway for motivating and regulating action. The ventral corticolimbic pathway evolved from the perirhinal cortex of the temporal lobe, closely connected with the insula, amygdala, and orbital frontal region (Pandya et al., 1982). In addition to supporting representation of objects or individual items in memory (Mishkin, 1982; Aggelton & Brown, 1999), the ventral pathway

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

supports action regulation with strong feedback control, in which sensory guidance operates to restrict the action plan to achieve the desired target (Goldberg, 1987; Passingham, 1987; Shima & Tanji, 1998). Recordings of cells from the arcuate premotor area, the ventrolateral frontal homolog of the mediodorsal SMA (Barbas & Pandya, 1986) show cells that are preferentially activated when actions are guided by visual cues (Tanji & Shima, 1996). An important theoretical challenge is to link the cybernetic qualities (projectional vs feedback control) to the cognitive representational qualities (configural versus object control). We argue that expectancies provide the links. Expectancies can be seen as consequences of the integration of the unique cybernetics of action regulation with the essential resources of memory representation to guide the cognitive process. At the limbic level, the representations of motive significance shape expectancies for goal attainment, and these representations must interact with the unfolding action plan recursively, as the motives give direction to the plans and the plans are evaluated for their goal relevance. Both dorsal and ventral corticolimbic pathways appear to contribute differently to the formation of expectancies (Tucker & Luu, 2007). The dorsal region of the anterior cingulate cortex is important to the general dorsal corticolimbic network involved in the representation of context-generated expectancies. The representation of a contextual map appears to have evolved to include the representation of reward expectancies as an integral component of the memory operation. That is, within an appropriate context, goal-directed actions can be internally generated independently of external input. This ballistic direction of action is supported by the expectancy for hedonic outcomes (Tucker & Luu, 2007). Functional neuroimaging studies show that the anterior cingulate cortex is particularly active when subjects must generate hypotheses (i.e., expectancies) about appropriate actions. In this light, akinetic mutism can be interpreted as reflecting an expectancy deficit, in that with no hedonic projection, there is no action. In everyday situations, the mediodorsal limbic-motor system appears to generate hypotheses that guide the launching and learning of appropriate actions, including communications with others. We theorize that the cybernetics of action regulation in the dorsal and ventral pathways maintains continuity with the unique motivational base of action in each pathway. The feedforward projectional control of action in the dorsal pathway is not only guided by the cognitive representation of configural relations (with the hippocampal support of spatial memory as the iconic exemplar); it also entails an inherent motivational bias toward hedonic expectancy that is consistent with launching goal-directed actions. This integrated operation of the brain thus supports a motivememory, not a neutral or disembodied cognitive function. As a result, it may be that in human cognition the representation of the current behavioral context within the dorsal pathway entails a positive hedonic tone to thoughts and actions.

 Don M. Tucker, Phan Luu & Catherine Poulsen

In the ventrolateral system, the amygdala is involved in forming object (i.e., cue)outcome associations, grounded in feedback control by viscerosensory constraints represented in insular cortex. This function supports the role of the adjacent orbitorfrontal cortex in representations of object expectancies (reward or punishment, Schoenbaum & Roesch, 2005) that guide actions. Likely, the representation of object-based expectancies provides required support for the ventrolateral prefrontal cortex involvement in rapid acquisition of arbitrary and abstract cues with actions (Bussey, Wise, & Murray, 2002). That is, with the ability to form reward and punishment expectancies for arbitrary cues, these cues can now motivate actions in a manner consistent with feedback control.

5. Aphasic disorders of action plans Aphasia syndromes provide important clues to the subcomponents of language, including the capacities required for hierarchic organization of grammatical structures. We propose that the mechanisms of language, as revealed by the aphasias, are fully interdependent with the mechanisms of action regulation (Tucker et al., 2008). The specificity of grammatical deficits with lesions to Broca’s area implies that this ventral corticolimbic network, at the base of the ventrolateral frontal motor system, is critical to complexity of clause structure. Several features of the ventral pathway, including the inhibitory control of semantic objects and the capacity for automatized action sequences, are critical to language generally and complex clause structure specifically. Even more generally, the cybernetics of the ventral trend may suggest new insight into the left hemisphere specialization for language, which we view as fundamentally a specialization of the entire hemisphere for the processing strategies of the ventral pathway (Liotti & Tucker, 1994; Tucker et al., 2008). At the same time as we emphasize the primacy of the ventral pathway for object memory and feedback control of actions, it should be apparent that the hierarchic organization of language structure, and its interpretation, require general cognitive skills requiring multiple brain networks. We suggest that an analysis of the unique memory mechanisms of both dorsal and ventral corticolimbic pathways, integrating both anterior motor and posterior sensory controls, is necessary for a full account of the process of organizing complex linguistic patterns.

5.1 Broca’s area: motor planning in the ventral pathway Lesions to a fairly restricted region of the brain, a few square centimeters in the caudal extent of the left inferior frontal lobe, result in deficits in language fluency. Because there are striking limitations in grammatical organization of speech, in contrast to relatively intact semantic reference, it is within Broca’s area that we must find critical

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

capacities for organizing grammatical complexity. To understand these capacities, we argue that it is necessary to appreciate the interdependence of this region of premotor (or perhaps pre-premotor) cortex with the memory capacities of the left temporal lobe. These memory capacities are supported by point-to-point reentrant connections through the uncinate fasciculus and they appear to be organized through recursive transactions across these connections. The result is to extend the unique inhibitory cybernetics of the object memory mechanisms of the ventral trend to create the powerful modular structures of language.

5.1.1 Inhibitory structure and the feedback control of action Within distributed neural networks such as make up the human brain, representational processes tend to engage the entire network, unless there are mechanisms for separating them. Concepts are therefore intrinsically holistic and syncretic, so that the separation of conceptual elements into chunks or packets — such as occurs with the clauses of speech — requires an active organizational mechanism. We suggest this mechanism is inhibition. To separate semantic units into interpretable bindings, the neural mechanisms of syntactic structure provide inhibitory control that is exerted by one representational unit (e.g., clause) on its associative neighbors. In typical language production, an important result of inhibitory specification of concepts may be the differentiation of the serial order of word production, aided by the routinized conventions of grammar, and carried out in the premotor networks at the ventral base of the left frontal lobe. When grammar is expert, its conventions allow complex organizations in which inhibitory control separates the meanings of subordinate clauses from superordinate ones, allowing scarce attention and working memory to be allocated to the superordinate level. Within this process, the culturally-trained automatization of meaning units within familiar clause structures provides sufficient memory capacity to hold the full structure of the utterance for an adequate interpretation. Lessons for the neural mechanisms producing this inhibitory specification of language may be provided by the mechanisms of action sequencing. Particularly important is the feedback form of action regulation within the ventral corticolimbic pathway. In contrast to the projectional control of the dorsomedial motor pathway, the feedback control of the ventrolateral motor pathway integrates perceptual checkpoints that are compared with the ongoing action sequence to allow sequential updates of the motor plan (Goldberg, 1985). This specification of the motor sequence requires inhibitory control that not only restricts the extent of each component of movement, but separates each component in relation to the sensory targets. In the reentrant connectivity from limbic to heteromodal to unimodal and primary cortices, the inhibitory specification in the ventral pathway appears to involve a unique form of recursion, in which

 Don M. Tucker, Phan Luu & Catherine Poulsen

the transformation of the representation with each processing cycle leads to greater differentiation and specification. To the extent that language production evolved from more generic communicative actions such as gestures (Givon, 1998), and to the extent that this evolutionary process required linguistic actions that are highly routinized, differentiated, and repeatable, it should not be surprising to find that it is specifically the ventral motor pathway and its capacity for inhibitory specification that has become the critical path for assembling intended meaning into linguistic structure.

5.1.2 Automatization of action and object structure With the specification of discrete actions within well-organized sequences, the ventral limbic-motor pathway is particularly suited to the development of routinized action patterns. Whereas the dorsal motor networks appear to embed actions within the episodic context that is elaborated within the configural representations of the posterior dorsal corticolimbic networks, the ventral motor networks appear to articulate more modular actions that are suited to serve as habitual patterns that can be disembedded from the immediate episodic context (Luu & Tucker, 2003; Tucker & Luu, 2007). In this way, the automatization of action in the anterior ventral networks is similar to the formation of perceptual objects in the posterior ventral networks, and it may rely on a similar mechanism of inhibitory specification. Objects are groupings of perceptual features that are sufficiently bound to be separated from the contextual frame. Routinized action packages are similar object representations of motor elements. It may be from the unique cybernetics of action objects in the ventral pathway that the patterns of grammar evolved in Broca’s and nearby networks. The patterns of grammar are automatized conventions of speech shared by a culture, allowing the members ready access to both the specific conventions and the more complex language structures that can be built on the foundation of these conventions.

5.1.3 Grammatical complexity in the motor pathway The position of grammatical complexity within the limbic-cortical pathway can be approached through developmental evidence. Grammar generally, and complex grammatical structure more particularly, are readily acquired by young children. But they are more difficult to acquire by second language learners after puberty, in contrast to basic lexical representations (agent, action, and object words). This differential learning capacity leads to pidgin constructions. This evidence can be interpreted to suggest that grammatical forms become rigid with the maturation of the neocortex of the motor pathways, which is relatively complete by the end of childhood. Even more fixed within motor neocortex are the prosodic and articulatory routines that allow native speech. Whereas grammar of a second language can be learned after puberty, speaking without an accent cannot. A similar fixity appears to hold for

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

the sensory networks of language comprehension, such that even if they learn the vocabulary of a new language rapidly, adults have great difficulty in “hearing” the unique sounds of a foreign language. The rationale for this reasoning about differential maturation in limbic versus neocortical networks comes from studies of maturation in the primate and human brain. It has long been apparent that a major sign of maturation, the increasing myelination of cortical fibers, occurs first in sensory and motor areas (Yakovlev & Lecours, 1967). More recently, studies of cortical anatomy have suggested that limbic cortical areas retain an immature biochemical compostion well into adulthood (Barbas, 2000). The implication is that plasticity is developmentally asymmetric between visceral-limbic and somatic-neocortical networks, such that by human adolescence there is rigidity in the somatic (neocortical) domain at the same time as there remains childlike plasticity in the visceral (limbic) domain. This neurodevelopmental perspective places grammatical complexity in an interesting position in the epistemology of human cognition. In contrast to lexical semantics, which remain plastic and dynamic into adulthood, grammar generally — and complexity specifically — become a mold for the mind, acquired automatically through experience with the culture of origin and quickly becoming resistant to later experience.

5.1.4 Left hemisphere specialization for the ventral trend The realization of the inhibitory representational cybernetics of the ventral pathway raises interesting questions for understanding a more well-known aspect of language localization in the brain, its left-lateralization. There is an integral role of object representations within the left hemisphere’s analytic cognitive capacities. This contrasts with the right hemisphere specialization for the spatial, configural concepts organized in the dorsal pathway. In light of the new understanding of dual corticolimbic representational systems, brain lateralization must be approached in a new light (Liotti & Tucker, 1994; Tucker, 2007). Not only do the right and left hemispheres’ perceptual skills align differently with the dorsal and ventral trends, respectively, but their motor capacities appear to do so as well. The ideomotor apraxias that are more common after left hemisphere lesions reflect not only a generic motor dominance of the right hand, but a precision of control that is commensurate with primary engagement of the inhibitory specification of action sequences in the ventrolateral premotor networks, elaborated particularly within the left hemisphere. Although there are of course both dorsal and ventral corticolimbic pathways within each cerebral hemisphere, there appears to be an asymmetry in the “dominance” of the archicortical (dorsal) and paleocortical (ventral) pathways within the right and left hemispheres, respectively. There may be a new way of looking at hemispheric specialization here, reflecting differential elaboration for one or the other of the

 Don M. Tucker, Phan Luu & Catherine Poulsen

corticolimbic pathways of perceptual integration, memory consolidation, and action regulation. For language, it is interesting to consider that many of the unique features that have been attributed to left hemisphere specialization may in fact represent more fundamental mechanisms of the object representations and sequence differentiation of the ventral corticolimbic pathway.

5.1.5 Restricted spreading activation and object memory In neural network models, and in neural networks, inhibitory control is critical to providing complex structures (Buzsaki, 2006). With only excitatory influences, interactions in the network are restricted to a kind of spreading activation, suitable for epileptic seizures but not for hierarchically organized neurocognitive processes. Spreading activation has become a useful model for understanding the cognitive mechanisms of semantic memory (Meyer, Osman, Irwin, & Yantis, 1988; Schvaneveldt & Meyer, 1976). Studies of reaction time have suggested that meaning spreads quickly and automatically from one word to related words. In making a word/nonword decision, subjects are faster to name words that have been “primed” by previous words that are semantically related (Meyer et al., 1988; Schvaneveldt & Meyer, 1976). Under an operative mechanism of spreading activation, there would be multiple meanings activated during the comprehension of a sentence, such that precision of meaning requires suppression of unintended associations (Gernsbacher & Faust, 1991; Gernsbacher & Kaschak, 2003). The left hemisphere may have special mechanisms for inhibitory specification of meaning, and we would argue that these mechanisms draw on the cybernetics of the ventral object memory pathway. Researchers have used right or left visual field (left or right hemisphere) presentation of prime and target words to examine whether spreading activation operates differently in the two hemispheres. Consistent with other evidence that the right hemisphere is important to comprehension of the gist or global meaning of language, some evidence has suggested that spreading activation appears broad and indiscriminate in the right hemisphere (Chiarello, 1985, 1988, 2000). In contrast, the spread of meaning is more restricted in the left hemisphere (Beeman, 1993; Beeman, Bowden, & Gernsbacher, 2000), consistent with greater inhibitory control of related meanings that are inappropriate to the immediate linguistic context. We speculate that these left and right hemispheric differences in spreading activation could reflect the more fundamental memory control biases of the ventral and dorsal corticolimbic pathways, respectively. With its specialization for the object memory and feedback control of the ventral limbic pathway, the left hemisphere gains a tight control over semantic structure, consistent with the inhibitory specification of actions that the recursive transformations in the ventral pathway appears to provide to motor control generally. With its specialization for the configural representation and feedforward mode of motor control of the dorsal corticolimbic pathway, the right hemisphere

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

gains a less constrained and more holistic structure of linguistic meaning that is suited to global comprehension of discourse and the implicit semantics of humor and allegory (Beeman et al., 2000).

5.2 Wernicke’s area: online self-monitoring Lesions of the posterior left hemisphere (Wernicke’s area) that lead to deficits of language comprehension do not simply impair comprehension. These lesions result in well-known expression deficits (jargon aphasia) in which grammatical form is correct, but semantic content is disordered (Goodglass, 1993). The presence of intact grammar with Wernicke’s aphasia is consistent with the argument that grammatical complexity is primarily a property of the motor preparatory networks of the frontal lobe. However, the interdependence of multiple networks in linguistic self-regulation is well illustrated by the deficits of Wernicke’s aphasia. The routinized cultural packets of verb and noun clauses have little meaning when they form automatically within inferior frontal networks and yet are unconstrained by semantic self-monitoring in the posterior receptive networks. The critical language networks of the frontal lobe are situated primarily within the ventral pathway. Similarly, those of the posterior temporal parietal (Wernicke’s) area must have considerable input from the ventral object memory pathway. However, it is an interesting question how much dorsal pathway input is integrated within Wernicke’s area (Galaburda & Pandya, 1983). Just as the parietal networks (dorsal pathway) are essential to motor control, apparently through providing dynamic monitoring of ongoing actions (Jeannerod, Arbib, Rizzolatti, & Sakata, 1995), there may be considerable integration of configural representations from the dorsal pathway as the posterior left hemisphere guides ongoing linguistic comprehension and expression.

5.3 Transcortical motor aphasia: inertia of language action A definite role for dorsal pathway control in language is shown by transcortical motor aphasia in which lesions of the mediodorsal regions of the frontal lobe lead to a paucity of spontaneous speech, even in the presence of intact articulatory capacity (Freedman, Alexander, & Naeser, 1984). This form of aphasia appears similar to akinetic mutism, with the motive deficit more specific to language processes. When questioned regarding their lack of spontaneous speech, transcortical motor aphasia patients often report that nothing comes to mind. This syndrome may thus reflect an impairment in the dorsal pathway’s normal contribution to the language process, which is a motivated, goal-oriented impulse to communicate. Lacking this normal feedforward extension of the visceromotor function, language is then directed only by the ventral pathway’s feedback control, such that speech is absent unless feedback direction is immediately present in the form of interpersonal confrontation.

 Don M. Tucker, Phan Luu & Catherine Poulsen

5.3.1 Alien speech Another clinical syndrome observed with lesions of the mediodorsal frontal lobe is the alien hand sign. The patient reports observing the actions of a hand, but not experiencing it as his/her own (Goldberg, Mayer, & Toglia, 1981). The implication of this disorder is that the motivational control of actions in the dorsal pathway is associated with an experience of the actions as integral to the self. Certainly we would expect that the consolidation of memory, arbitrating as it does between the somatic networks of sensory and motor neocortices and the visceral networks of the limbic cortices, would result in representations with both environmental veracity and personal motive significance. However, the clinical literature shows no counterpart to the loss of felt personal significance of actions with lesions to the ventral limbic-motor pathway. Instead, patients with lesions to orbital and ventrolateral frontal cortex often show behavioral disinhibition, puerile impulsivity, and indifference to social norms, in the pseudopsychopathic syndrome (Blumer & Benson, 1975). If language can be carried out more or less independently within the ventral pathway, but the experience of personal agency requires the participation of the dorsal pathway, it is interesting to consider the thought disorder of schizohrenia, in which internal speech is apparently divorced from the sense of personal agency, and is instead experienced as an alien intrusion into the mind (Bick & Kinsbourne, 1987).

6. Dialectical cybernetics of linguistic complexity We have theorized that there are unique neural mechanisms necessary for grammatical complexity, emergent from the capacity for inhibitory specification of objects from their embedding contexts that is achieved by the ventral corticolimbic networks of the left hemisphere. These mechanisms appear to build upon similar capacities for inhibitory specification of discrete and serial actions within the ventrolateral regions of the frontal lobe. These are fundamentally motor control processes, and yet they have critical implications for representation, allowing relational clauses to be bound as units, to be organized hierarchically within expressive or receptive sequences. At the same time, however, as the cybernetics of object memories are applying inhibitory specification to first differentiate and then maintain clausal structure, any hierarchic organization of the semantic context must draw on multiple brain systems. Although the patient with right hemisphere damage may appear to have intact language, more careful testing shows important limitations in understanding the implicit, connotative, and contextual referents in extended discourse (Borod, 2000). Apparently, the left hemisphere (and ventral corticolimbic) skills in differentiated, routinized semantics are balanced by the right hemisphere (and dorsal corticolimbic) skills in pragmatic representation of the adaptive context of the communication.

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

A general framework for the recursive processing across the reentrant connections of the four levels of corticolimbic network architecture has been suggested to be dialectical (Tucker, 2001; 2007), in that opposing demands of visceral (motive significance) and somatic (environmental interface) representations are arbitrated to achieve abstract concepts. In the more specific case of language, we can consider not only a general visceral-somatic dialectic process of memory consolidation, but unique contributions of ventral and dorsal corticolimbic pathways to this process, contributions which themselves are elaborated differently within the left and right hemispheres. Thus, the left hemisphere’s ventral frontal language networks are thus critical for language. Yet they normally operate in a balanced, perhaps dialectical, fashion, with other networks, with opposing control biases at one level leading to stability — and complexity — at a higher level. The left frontal organization of efficient grammatical constructions is continuously monitored by posterior networks to provide constraints of meaningfulness against which the construction proceeds effectively. The grammatical structure may begin as a primitive form, a motive urge, but it is then articulated to the specified structure of the utterance through recursive transformations in which inhibitory specification is a continuing constraint. Verbal objects are differentiated from the embedding semantic context within the left hemisphere’s ventral networks, and yet that context can be maintained on-line together with its configural implications, perhaps most clearly within the dorsal, pragmatic, networks of the right hemisphere. More fundamentally, the motive to communicate grounds the linguistic process in an adaptive context. The representation of self and other that frames that motive may be preferentially formed within the dorsal corticolimbic pathway that reflects the context of communication. In this way, the neural mechanisms of syntactic complexity can be seen as affording an efficiency of memory usage and cognitive representation that expands conscious access to the multiple streams of information processing that contribute to social communication.

References Aggleton, J.R. & Brown, M.W. 1999. Episodic memory, amnesia, and the hippocampal-anterior thalamic axis. Behavioral and Brain Sciences 22: 425–489. Allport, A. 1985. Selection for action: Some behavioral and neurophysiological considerations of attention and action. In Issues in Perception and Action, H. Heuer & A.F. Sanders (Eds), Hillsdale NJ: Lawrence Erlbaum. Baddeley, A.D. & Hitch, G. 1974. Working memory. In The Psychology of Learning and Motivation: Advances in Research and Theory, Vol. 8, G.H. Bower (Ed.), 47–89. New York NY: Academic Press. Barbas, H. 2000. Complementary roles of prefrontal cortical regions in cognition, memory, and emotion in primates. Adv Neurol 84: 87–110.

 Don M. Tucker, Phan Luu & Catherine Poulsen Barbas, H. & Pandya, D.N. 1986. Architecture and frontal cortical connections of the premotor cortex area 6 in the rhesus monkey. Journal of Comparative Neurology 256: 211–228. Barris, R.W. & Schuman, H.R. 1953. Bilateral anterior cingulate gyrus lesions. Neurology 3: 44–52. Beeman, M. 1993. Semantic processing in the right hemisphere may contribute to drawing inferences from discourse. Brain Lang 44(1): 80–120. Beeman, M.J., Bowden, E.M. & Gernsbacher, M.A. 2000. Right and left hemisphere cooperation for drawing predictive and coherence inferences during normal story comprehension. Brain Lang 71(2): 310–336. Bick, P.A. & Kinsbourne, M. 1987. Auditory hallucinations and subvocal speech in schizophrenic patients. American Journal of Psychiatry 1(44–1): 222–225. Blumer, D. & Benson, D.F. 1975. Personality changes with frontal and temporal lobe lesions. In Psychiatric Aspects of Neurologic Disease, D.F. Benson & D. Blumer (Eds), 151–170. New York NY: Gruen and Stratton. Borod, J. 2000. The Neuropsychology of Emotion. Oxford: OUP. Brown, J. 1977. Mind, Brain, and Consciousness: The Neuropsychology of Cognition. New York NY: Academic Press. Brown, J.W. 1988. The Life of the Mind: Selected Papers. Hillsdale NJ: Lawrence Erlbaum. Bussey, T.J., Wise, S.P. & Murray, E.A. 2002. Interaction of ventral and orbital prefrontal cortex with inferotemporal cortex in conditional visuomotor learning. Behav Neurosci 11(64): 703–715. Buzsaki, G. 2006. Rhythms of the Brain. Oxford: OUP. Chiarello, C. 1985. Hemisphere dynamics in lexical access: Automatic and controlled priming. Brain Lang 2(61): 146–172. Chiarello, C. 1988. Lateralization of lexical processes in the normal brain: A review of visual half-field research. In Contemporary Reviews in Neuropsychology, H. Whitaker (Ed.), 36–76. New York NY: Springer. Chiarello, C. 2000. Back to the future: Diversity and coherence in brain and language studies. Brain Lang 7(11): 42–43. Crosson, B. 1999. Subcortical mechanisms in language: Lexical-semantic mechanisms and the thalamus. Brain Cogn 4(02): 414–438. Daneman, M. & Carpenter, P.A. 1980. Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior 19: 450–466. Freedman, M., Alexander, M.P. & Naeser, M.A. 1984. Anatomic basis of transcortical motor aphasia. Neurology 34: 409–417. Gabriel, M., Lambert, R.W., Foster, K., Orona, E., Sparenborg, S., & Maiorca, R.R. 1983. Anterior thalamic lesions and neuronal activity in the cingulate and retrosplenial cortices during discriminative avoidance behavior in rabbits. Behav Neurosci 9(75): 675–696. Gabriel, M., Sparenborg, S. & Kubota, Y. 1989. Anterior and medial thalamic lesions, discriminative avoidance learning, and cingulate cortical neuronal activity in rabbits. Exp Brain Res 76(2): 441–457. Galaburda, A.M. & Pandya, D.N. 1983. The intrinsic architectonic and connectional organization of the superior temporal region of the rhesus monkey. J Comp Neurol 221(2): 169–184. Gernsbacher, M.A. & Faust, M.E. 1991. The mechanism of suppression: A component of general comprehension skill. J Exp Psychol Learn Mem Cogn 172: 245–262. Gernsbacher, M.A. & Kaschak, M.P. 2003. Neuroimaging studies of language production and comprehension. Annu Rev Psychol, 541, 91–114.

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

Gibson, E. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition 68(1): 1–76. Givón, T. 1998. On the co-evolution of language, mind, and brain. Institute of Cognitive and Decision Sciences TR-98–07. Goldberg, G. 1985. Supplementary motor area structure and function: Review and hypotheses. Behavioral and Brain Sciences 8: 567–616. Goldberg, G., Mayer, N.H. & Toglia, J.U. 1981. Medial frontal cortex infarction and the alien hand sign. Arch Neurol 38(11): 683–686. Goodglass, H. 1993. Understanding aphasia. New York NY: Academic Press. Guillery, R.W. & Sherman, S.M. 2002. Thalamic relay functions and their role in corticocortical communication: Generalizations from the visual system. Neuron 33(2): 163–175. Halford, G.S., Baker, R., McCredden, J.E. & Bain, J.D. 2005. How many variables can humans process? Psychol Sci 16(1): 70–76. Halford, G.S., Cowan, N. & Andrews, G. 2007. Separating cognitive capacity from knowledge: A new hypothesis. Trends Cogn Sci 11(6): 236–242. Hendler, J.A. 1995. Types of planning: Can artificial intelligence yield insights into prefrontal function? In Structure and Functions of the Human Prefrontal Cortex, Vol. 769, J. Grafman, K.J. Holyoak & F. Boller (Eds), 265–276. New York NY: New York Academy of Sciences. Hannus, A., Cornelissen, F.W., Lindemann, O. & Bekkering, H. 2005. Selection-for-action in visual search. Acta Psychol Amst 118(1–2): 171–191. Hauser, M., Chomsky, N. & Fitch, W.T. 2002. The faculty of language. What is it, who has it, and how did it evolve? Science 298: 1569–1579. Jackson, J.H. 1931. The evolution and dissolution of the nervous system. In Selected Writings of John Hughlings Jackson, Vol. II, 45–75. London: Hodder and Stoughton. Jeannerod, M., Arbib, M.A., Rizzolatti, G. & Sakata, H. 1995. Grasping objects: The cortical mechanisms of visuomotor transformation. Trends in Neuroscience 18: 314–320. Just, M.A. & Varma, S. 2007. The organization of thinking: What functional brain imaging reveals about the neuroarchitecture of complex cognition. Cogn Affect Behav Neurosci 7(3): 153–191. Karlsson, F. 2007. Constraints on multiple center-embedding of clauses. Journal of Linguistics 43: 365–392. Keele, S.W. 1981. Behavioral analysis of movement. Handbook of Physiology: Section 1: The Nervous System, Vol. II, Motor Control Part 2, 31: 1391–1413.0. Keele, S.W. & Hawkins, H.H. 1982. Explorations of individual differences relevant to high level skill. Jn. Motor Beh 14: 3–23. King, J. & Just, M.A. 1991. Individual differences in syntactic processing: The role of working memory. Journal of Memory and Language 30: 580–602. Herrick, C.J. 1948. The Brain of the Tiger Salamander. Chicago IL: University of Chicago Press. Lashley, K.S. 1951. Cerebral Mechanisms in Behavior. New York NY: Wiley. Liotti, M. & Tucker, D.M. 1994. Emotion in asymmetric corticolimbic networks. In Human Brain Laterality, R.J. Davidson & K. Hugdahl (Eds), 389–424. Oxford: OUP. Luu, P., Flaisch, T. & Tucker, D.M. 2000. Medial frontal cortex in action monitoring. J Neurosci 20(1): 464–469. Luu, P. & Tucker, D.M. 1998. Vertical integration of neurolinguistic mechanisms. In Handbook of Neurolinguistics, B. Stemmer & H.A. Whitaker (Eds), 55–203. San Diego CA: Academic Press.

 Don M. Tucker, Phan Luu & Catherine Poulsen Luu, P. & Tucker, D.M. 2003. Self-regulation and the executive functions: Electrophysiological clues. In The Cognitive Electrophysiology of Mind and Brain, A. Zani & A.M. Preverbio (Eds), 199–223. San Diego CA: Academic Press. Mesulam, M.M. 2000. Behavioral neuroanatomy: Large-scale networks, association, cortex, frontal syndromes, the limbic system, and hemispheric specializations. In Principles of behavioral and cognitive neurology, M.M. Mesulam (Ed.),1–120. Oxford: OUP. Meyer, D.E., Osman, A.M., Irwin, D.E. & Yantis, S. 1988. Modern mental chronometry. Biol Psychol, 2(61–3): 3–67. Miller, G.A. 1956. The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review 63: 81–97. Mishkin, M. 1982. A memory system in the monkey. Philos Trans R Soc Lond B Biol Sci 298(1089): 83–95. Nadel, L. 1991. The hippocampus and space revisited. Hippocampus 1: 221–229. Neafsey, E.J., Terreberry, R.R., Hurley, K.M., Ruit, K.G. & Frysztak, R.J. 1993. Anterior cingulate cortex in rodents: Connections, visceral control functions, and implications for emotion. In Neurobiology of the Cingulate Cortex and Limbic Thalamus, B.A. Vogt & M. Gabriel (Eds), 206–223. Boston MA: Birkhauser. Pandya, D.N. & Seltzer, B. 1982. Association areas of the cerebral cortex. Trends in Neural Science 5: 386–390. Pandya, D.N. & Barnes, C.L. 1987. Architecture and connections of the frontal lobe. In The Frontal Lobes Revisited, E. Perecman (Ed.), 41–72. New York NY: IRBN. Pandya, D.N. & Seltzer, B. 1982. Association areas of the cerebral cortex. Trends in Neural Science 5: 386–390. Pandya, D.N. & Yeterian, E.H. 1984. Proposed neural circuitry for spatial memory in the primate brain. Neuropsychologia 222: 109–122. Papez, J.W. 1937. A proposed mechanism of emotion. Archives of Neurology and Psychiatry 38: 725–743. Passingham, R.E. 1987. Two cortical systems for directing movement. In Motor Areas of the Cerebral Cortex, C. Foundation (Ed.), 151–164. New York NY: John Wiley and sons. Prat, C.S., Keller, T.A. & Just, M.A. 2007. Individual differences in sentence comprehension: A functional magnetic resonance imaging investigation of syntactic and lexical processing demands. J Cogn Neurosci 19: 1950–1963. Pribram, K.H. 1991. Brain and Perception: Holonomy and Structure in Figural Processing. Hillsdale NJ: Lawrence Erlbaum. Schneider, W. & Chein, J. 2003. Controlled and automatic processing: Behavior, theory, and biological mechanisms. Cognitive Science 27: 525–559. Schneider, W. & Shiffrin, R.M. 1977. Controlled and autonomic human information processing: I. Detection search and attention. Psychological Review 84(1): 1–66. Schvaneveldt, R.W. & Meyer, D.E. 1976. Lexical ambiguity, semantic context, and visual word recognition. J Exp Psychol Hum Percept Perform 22: 243–256. Sherman, S.M. & Guillery, R.W. 2002. The role of the thalamus in the flow of information to the cortex. Philos Trans R Soc Lond B Biol Sci 357(1428): 1695–1708. Shiffrin, R.M. & Schneider, W. 1977. Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review 84(1): 127–190. Shima, K. & Tanji, J. 1998. Both supplementary and presupplementary motor areas are crucial for the temporal organization of multiple movements. J Neurophysiol 80(6): 3247–3260.

Neural mechanisms of recursive processing in cognitive and linguistic complexity 

Squire, L.R. 1986. Mechanisms of memory. Science, 232, 1612–1619. Squire, L.R. 1998. Memory systems. C R Acad Sci III 321(2–3): 153–156. Tanji, J. & Shima, K. 1996. Supplementary motor cortex in organization of movement. Eur Neurol 36 Suppl 1, 13–19. Tucker, D.M. 2001. Motivated anatomy: A core-and-shell model of corticolimbic architecture. In Handbook of Neuropsychology, 2nd Edn. Vol. 5, Emotional Behavior and its Disorders, G. Gainotti (Ed.), Amsterdam: Elsevier. Tucker, D.M. 2007. Mind From Body: Experience From Neural Structure. Oxford: OUP. Tucker, D.M., Frishkoff, G.A. & Luu, P. 2008. Microgenesis of language: Vertical integration of neurolinguistic mechanisms across the neuraxis. In Handbook of the Neuroscience of Language, B. Stemmer & H.A. Whitaker (Eds), Oxford: OUP. Tucker, D.M. & Luu, P. 2006. Adaptive binding. In Binding in Human Memory: A Neurocognitive Approach, H. Zimmer, A. Mecklinger & U. Lindenberger (Eds), 85–108. Oxford: OUP. Tucker, D.M. & Luu, P. 2007. Neurophysiology of motivated learning: Adaptive mechanisms of cognitive bias in depression. Cognitive Therapy and Research 31(2): 189–209. Yakovlev, P.I. 1948. Motility, behavior and the brain. The Journal of Nervous and Mental Disease 107: 313–335.

Syntactic complexity in the brain Angela D. Friederici & Jens Brauer

Max Plance Institute for Human Cognitive and Brain Sciences The current chapter considers neuronal circuits in the human brain that represent a neuroanatomical basis for the processing of syntactic complexities. We will present data from event-related brain potential studies and from functional and structural brain imaging studies to elucidate the brain’s underpinnings for syntactic processing. The data shall indicate that the processing of syntactic dependencies is subserved by two distinct networks of brain areas, one involving the deep frontal operculum and the anterior part of the superior temporal gyrus (STG), holding responsible for the processing of local dependencies, the other involving Broca’s area and the posterior part of the STG, holding responsible for the processing of hierarchical dependencies. Structural brain data are referred that identify two separate neural fiber pathways for these two networks. These findings are supported by ontogenetic and phylogenetic comparison. The data suggest functional and structural separation for the processing of different levels of syntactic complexity.

1. Introduction The human capacity of processing language and syntactic operations still remains an unsolved issue. There is agreement, however, that the ability to process complex, hierarchically structured sentences is specific to humans. Several views on the acquisition of language and syntax and underlying brain mechanisms have been proposed. These views ranged from the emergence of elementary linguistic universals by biological adaptation through natural selection (Pinker & Bloom, 1990) whereby a learned feature finally becomes an intrinsic trait resembling an instinct (Baldwin effect). On the other hand there are views relying on a broader basis for language and communication skills such as social-cognitive and social-motivational structures (Tomasello, 2007) or evolution on the basis of non-linguistic constraints and their termination in a linguistic communication system (see Christiansen & Kirby, 2003). All these views acknowledge the brain’s relevance to the evolution of language, either with respect to domain-specialized brain systems or with respect to its domain-general processing constraints. The following considerations will discuss in more detail ontogenetic and phylogenetic developments of the brain basis for processing syntactic complexities.

 Angela D. Friederici & Jens Brauer

Behavioral studies in non-human primates provided data that these animals were able to learn a Finite State Grammar (FSG) of the (AB)n type, but not to learn a simple Phrase Structure Grammar (PSG) of the AnBn type (Fitch & Hauser, 2004). This is in contrast to humans who learn both classes of grammar equivalently well. From these data the conclusion of a specific human capacity to process hierarchical structures was drawn. This interpretation and the data were seen to support the claim that the ability to process recursive structures is the very narrow aspect that distinguishes human from non-human communication systems (Hauser, Chomsky, & Fitch, 2002). Though, obviously, the empirical support at this time was only preliminary since the study cited above employed a PSG that made use of hierarchical and not necessarily recursive structures. Beyond stimulating comparative behavioral studies, the argument put forward by Hauser et al. (2002) also led to considerations that, if there was a principled difference between human and non-human primates, this might manifest in a specialized brain system fully developed in the human brain only (Friederici, 2004b). As suggested by the existing literature on syntax processing in humans, local syntactic dependencies such as the (AB)n type involved the ventral premotor cortex (vPMC, Brodmann area (BA) 6) and the deep frontal operculum (FO), while complex syntactic processes were subserved by Broca’s area (BA 44/45). A differentiation between these cortical areas is supported by some theories on brain evolution with vPMC/FO being phylogenetically older than Broca’s area (Sanides, 1962). Converging evidence on distinct characteristics between vPMC/FO and Broca’s area also stems from cytoarchitectonic studies classifying cortical areas according to their cell layer structure (Amunts, et al., 1999). Thereby, the existence and quantity of granular cells in cortical layer IV distinguishes BA 44 being dysgranular cortex and BA 45 being granular cortex from agranular cortex in vPMC. This differentiation between BA 44/45 and vPMC was already utilized by Brodmann in his original work (Brodmann, 1909). The way to robust decisions about homologies and non-homologies between human and non-human brains is a tricky business (Arbib & Bota, 2003). Nevertheless, there is apparent agreement that when comparing across species, human BA 44 seems to resemble what has been described in the non-human primate (macaque monkey) brain as area 44 (Petrides & Pandya, 1994) or area F5 (Matelli, Luppino, & Rizzolatti, 1985) though human BA 44 and macaque area 44 differ cytoarchitectonically. While human BA 44 is dysgranular, area 44 in macaques is agranular, comparable to human BA 6. This is resembled by functional involvement. Macaque agranular area 44 is concerned with orofacial movement like human BA 6 (Petrides, Cadoret, & Mackey, 2005), while human dysgranular BA 44 is particularly involved in processing syntactic operations (Friederici, 2004a). Comparison between human and macaque cortex suggests that relative size of BA 44 (in relation to BA 6 and BA 45) is much smaller in the macaque than in the human brain (Petrides & Pandya, 1994). Unfortunately, comparisons to our closest non-human primate relatives, chimpanzees and bonobos, are not available so far.

Syntactic complexity in the brain 

An alternative and extended view on brain structure is provided by an investigation of subcortical connectivity between cortex areas, in the present context particularly those regions that participate in syntactic processes. A review of the current literature revealed that processing of syntactically complex sentences is supported by BA 44/45 together with the posterior part of the superior temporal gyrus, STG (Grodzinsky & Friederici, 2006). The dominant fiber pathway between these areas is the superior longitudinal fasciculus (SLF), a dorsal connection between the prefrontal and the temporal cortical regions (Catani, Jones, Donato, & ffytche, 2003; Friederici, Bahlmann, Heim, Schubotz, & Anwander, 2006). In the macaque brain, the SLF connects prefrontal with parietal regions and does not extent into the temporal region. Rather, the main prefrontal-temporal connection in the macaque brain is provided by a ventral connection that includes the extreme capsule (Schmahmann, et al., 2007). This ventral pathway is existent in humans, too, where it was shown to connect the temporal region to the FO and being involved in the processing of non-complex, local syntactic dependencies (Friederici, Bahlmann, et al., 2006).

2. Two processing stages A comparative investigation of structural brain connectivities between humans, chimpanzees, and macaques revealed divergence across species (Rilling, et al., 2008). In the human brain, strong dorsal connections associate prefrontal areas BA 44, 45, and 47 to the temporal region, while these connections where much weaker and shorter in the chimpanzee brain. In the macaque brain, moreover, the ventral association was dominant. This result was seen to support the view of a phylogenetic evolvement of the dorsal fronto-temporal connection, probably related to the evolution of language functions. Ontogenetic brain development is characterized by some major changes and adaptations in brain morphometry, but also by concurrent growing sensory, motor, executive, and cognitive functions, among them increasing skills in syntactic processing. Structural maturation of fiber pathways in the brain is particularly characterized by increasing myelination of the fibers. Myelination refers to formation and progression of myelin sheaths which originate from extended plasma membranes of oligodendrocytes wrapping around segments of an axon. Thereby, the speed of electric impulses along the axon is increased and hence the speed and accuracy of information exchange in neuronal networks as well. During early childhood, the volume increase in myelinated white matter shows a very similar time course in frontal and temporal (i.e., language-related) regions which is, moreover, later in time than the volume increase in the (sensorimotor) central region (Pujol, et al., 2006). Simultaneously, processing skills on syntactic complexities advance, such as the processing of passive sentences (Fox & Grodzinsky, 1998) or case marked non-canonical object-first sentences (Dittmar, Abbot-Smith, Lieven, & Tomasello, 2008).

 Angela D. Friederici & Jens Brauer

The fronto-temporal fiber connections, particularly the SLF, which appears the crucial pathway connection in the context of complex syntactic processing, are the very late fiber tracts to mature. They reach adult-like myelination only during late adolescence and young adulthood (Giorgio, et al., 2008; Lebel, Walker, Leemans, Phillips, & Beaulieu, 2008). As outlined above, on the basis of empirical evidence syntactic processes were proposed to rely on two different neuronal networks in the brain (Friederici, Bahlmann, et al., 2006; Grodzinsky & Friederici, 2006). One network consists of the FO and the anterior STG and is involved in constructing local phrase structures. The other one consists of BA 44/45 and the posterior STG and comes additionally into play when more complex dependency relations between constituents of a sentence (complex syntax) are to be processed. Starting from theoretical considerations that parsing involves two processing stages, a first processing stage during which local phrase structures are built-up on the basis of word category information and a second stage during which grammatical relations between constituents are assigned (Frazier, 1987), a number of experiments evaluated these different processing stages (see Friederici, 2002).

3. Local phrase structure circuit The first experiments on this framework focused on local phrase structure building representing the first processing stage. Event-related brain potentials (ERPs) were measured with a violation paradigm in which the processing of correct sentences was compared to the processing of sentences containing a phrase structure violation, i.e., a violation of the obligatorily required word category. Two ERP components in response to a phrase structure violation were observed, an early left anterior negativity (ELAN) and a late centro-parietal positivity around 600 ms (P600) (Friederici, 1995; Friederici, Hahne, & Mecklinger, 1996; Hahne & Friederici, 1999). The P600 component had been observed in response to a number of different syntactic anomalies including syntactic violations and syntactic ambiguities (Hagoort, 1993; Osterhout & Holcomb, 1993; Osterhout, Holcomb, & Swinney, 1994) as well as syntactically complex sentences (Kaan, Harris, Gibson, & Holcomb, 2000), and is widely taken to reflect syntactic processing. Therefore, the P600 was interpreted to reflect a late processing phase during which the ultimate syntactic relations are assigned and during which, if necessary, syntactic revision and repair takes place. The ELAN which occurs with outright syntactic phrase structure violations was taken to reflect an initial processing phase of local phrase structure built-up, i.e., the inability to build a local phrase structure due to an element with the incorrect word category (e.g., verb instead of noun; correct: The pizza was in the fridge vs. incorrect: The

Syntactic complexity in the brain 

pizza was in the eaten). The ELAN effect was reported in a number of studies and languages (Friederici, Pfeifer, & Hahne, 1993; Neville, Nicol, Barss, Forster, & Garrettdew, 1991; Ye, Luo, Friederici, & Zhou, 2006). Dipole localization analysis of this early syntactic effect in magnetoencephalography (MEG) data revealed two dipoles in each hemisphere with maxima in the left hemisphere: one in the anterior STG and one in the inferior frontal region (Friederici, Wang, Herrmann, Maess, & Oertel, 2000). The involvement of the anterior temporal region and the inferior frontal region during the process reflected in the ELAN was confirmed by lesion data. In patients with lesions in the anterior temporal lobe, the ELAN component was absent, and the same was true for patients with lesions in the inferior frontal cortex (Friederici, Hahne, & von Cramon, 1998; Friederici & Kotz, 2003; Friederici, von Cramon, & Kotz, 1999). Patient data additionally revealed that only left inferior frontal cortical lesions but not left subcortical lesions in the basal ganglia lead to an absence of the ELAN indicating that the left anterior cortical structures are relevant for the processes reflected by the ELAN (Friederici, et al., 1999). With respect to their developmental unfolding during ontogeny, the presence of these syntax-related ERP components clearly depends on the sentence structure in which the syntactic violations are realized, and, moreover, on what type of violation is implemented. In simple age-appropriate active sentences employing a word category violation (correct: The lion roars vs. incorrect: The lion in the roars), syntactic ERP components are observable already in young children in the 3rd year of life: an ELANP600-like pattern that, however, shows longer latencies and sustained duration in young children (Oberecker, Friedrich, & Friederici, 2005). On the other hand, 2-year-olds do only show a late positivity but not the early negativity, when confronted with the same material (Oberecker & Friederici, 2006) (see Figure 1). A late positivity was also reported for syntactic processes elicited by morphosyntactic violations in 3 and 4-year-olds (Silva-Pereyra, Rivera-Gaxiola, & Kuhl, 2005). Comparably, but shifted in age, more complex sentence structure such as passive constructions (see above) elicit in children ELAN-P600-like effects only at age 7 and older, while 6-year-olds merely display a P600like effect but no early negativity (Hahne, Eckstein, & Friederici, 2004) (see Figure 2). These data show that the child’s brain is sensitive to phrase structures. They also show that both ERP components are independent from each other, suggesting that they reflect two separate processes. It is, moreover, noteworthy to keep in mind that the child’s brain response depends on both, the age of the child (and hence the brain’s maturity and functional capacity), but also the complexity of the syntactic structures that are to be processed. Early automatic syntactic processes reflected in the ELAN are acquired ontogenetically later than the more controlled processes reflected in the P600. An fMRI study in adults and 6-year-old children was conducted with analogous ageappropriate sentence material in active voice to specify in more detail the brain regions that are involved (Brauer & Friederici, 2007). The data for adults showed increased

 Angela D. Friederici & Jens Brauer 24-month-olds -10 mV

10

32-month-olds

F7

s 0.4 0.8 1.2 1.6

-10 mV ELAN

10

Adults -5 mV ELAN

F7

s 0.4 0.8 1.2 1.6

PZ

F7

s 0.4 0.8 1.2 1.6

5

PZ

P600

F7 PZ

PZ

P600

syntactically correct incorrect

P600

Figure 1. Event-related potentials (ERPs) from electroencephalographic (EEG) data in 24-month-olds, 32-month-olds, and adults in response to auditory presentation of syntactically incorrect and correct sentences in active voice. Adults show a brain response with an early left anterior negativity (ELAN) and a late positivity (P600). This typical pattern is also observed in 32-month-olds, but not yet in 24-month olds who only show the P600. The very early automatic structure building processes are not yet developed in this age group. Figure modified from Oberecker & Friederici (2006).

6-year-olds -10 mV

10

10-year-olds -10 mV ELAN

F7 s 0.5

1.0

1.5

10

-5 mV

F7 s

0.5

PZ

P600

Adults

1.0

1.5

5

PZ

P600

ELAN F7 s 0.5

1.0

F7 PZ

1.5 PZ

syntactically correct incorrect

P600

Figure 2. ERP data from different age groups of children and from adults during the processing of syntactically incorrect and correct sentences. Unlike the data in Figure 1, the sentence material used for the data in Figure 2 are syntactically more complex (passive voice). 6-year-old children do not yet display an ELAN effect on syntactically incorrect sentences but they do already show a late positivity. Only children older than 6 years of age show an adult-like pattern with an early anterior negativity and a late posterior positivity. Figure modified from Hahne, et al. (2004).

Syntactic complexity in the brain 

activation in the STG and in the anterior FO for syntactically incorrect sentences compared to correct and to semantically incorrect sentences. In the STG, particularly its posterior part showed enhanced activation compared to the processing of correct sentences. This region was also stronger active during the processing of semantically incorrect sentences, suggesting its involvement in syntactic and semantic sentential processes. While the midportion of the STG showed increased activation for both violation types, the anterior STG and the anterior FO displayed more activation for sentences with syntactic phrase structure violations compared to correct sentences and to sentences with semantic violations. Thus, it appears that the FO, anterior STG and posterior STG are supporting syntactic processes. Based on the MEG localization data for the ELAN effect which showed an involvement of the anterior STG and the frontal brain region and by logical exclusion, one might hypothesize that the posterior STG supports processes reflected by the P600. In the same study, children, in contrast to adults, activated the entire STG and also the FO in all conditions. But they, unlike adults, distinguished the syntactic condition by an increased involvement of Broca’s area. In adults, Broca’s area was only involved at a low threshold (Brauer, Neumann, & Friederici, 2008). The stronger involvement of Broca’s area in children is likely to reflect higher processing demands in the developing system for syntactic processes. Such stronger involvement of Broca’s area is observed in adults during the processing of higher syntactic complexity or longer syntactic dependencies (Bornkessel, Zysset, Friederici, von Cramon, & Schlesewsky, 2005; Friederici, Fiebach, Schlesewsky, Bornkessel, & von Cramon, 2006). For example, in one of these studies, the inferior portion of BA 44 showed parametrically stronger activation with increasing syntactic complexity operationalized as the number of objects moved in front of the subject (Friederici, Fiebach, et al., 2006). In earlier studies with non-parametric designs, Broca’s area (among other activation sites) had been reported to show stronger activation for syntactically more complex compared to less complex sentences in a variety of languages: German (Roder, Stock, Bien, Neville, & Rosler, 2002), English (Stromswold, Caplan, Alpert, & Rauch, 1996), Hebrew (Ben-Shachar, Hendler, Kahn, Ben-Bashat, & Grodzinsky, 2003). A conclusion from these fMRI studies might be that Broca’s area is recruited when complex and more demanding syntactic structures are processed, whereas local phrase structure building does not necessarily recruit this area, but rather a neighboring area deep inside the inferior frontal cortex, i.e., the frontal operculum which is by some models of brain evolution regarded a phylogenetically older brain region.

4. Complex syntax circuit Studies that directly contrasted local syntactic processes and complex syntactic processes further elucidated the functional specification of the relevant brain areas. In

 Angela D. Friederici & Jens Brauer A

Finite state grammar (AB)n

A

B

A

B

A

B

Phrase structure grammar AnBn

A

B

cor/short: A B A B de bo gi fo viol/short: A B A A de bo gi le cor/long: A B A B A B A B le ku ri tu ne wo ti mo viol/long: A B A B A B A A le ku ri tu ne wo ti se

A A AA

B B B B

cor/short: A A B B ti le mo gu viol/short: A A B A ti le mo de cor/long: A A A A B B B B le ri se de ku bo fo tu viol/long: A A A A B B B A le ri se de ku bo fo gi

syllables of category A: de, gi, le, mi, ne, ri, se, ti syllables of category B: bo, fo, gu, ku, mo, pu, to, wu B

Phrase structure grammar AnBn

Finite state grammar (AB)n

A1

B1

A2

B2

A3

B3

cor/short: A2 B2 A3 B3 de to gi ko viol/short: A2 B2 A3 B1 de tu ge pu cor/long: A1 B1 A3 B3 A2 B2 be pu gi ku de to viol/long: A3 B3 A1 B1 A2 B3 ge ku bi po di ko

A1 A2 A3

B3 B2 B1

cor/short: A1 A2 B2 B1 viol/short: A1 A2 B2 B3 cor/long: A3 A1 A2 B2 B1 B3 viol/long: A3 A1 A2 B2 B1 B2

bi de to pu be de tu ku ge bi di tu po ko ge bi di tu po to

syllables of category A: be, bi, de, di, ge, gi syllables of category B: po, pu, to, tu, ko, ku Relation between An−Bn: voiced - unvoiced consonant

Figure 3. Illustration and examples for two grammar rules: finite state grammar (FSG) and phrase structure grammar (PSG). Figure 3A exemplifies the two grammar rules as implemented in Friederici, Bahlmann, et al. (2006). The FSG rule was generated by straightforward transitions between categories of consonant-vowel syllables. The PSG rule was generated by embeddings between the two syllable categories. Both grammars were utilized with short and long sequences. Violations of the rule structure were located at the last 3 or 4 positions (short sequences) and at the last 4, 5, or 6 positions (long sequences). In the example given in 3A, the violations are located at the fourths position for short sequences and at the sixths position for long sequences (see bold letters). Figure 3B exemplifies the two grammar rules as implemented in Bahlmann, et al. (2008). Categories members (A and B) were coded phonologically with category “A” syllables containing the vowels “i” or “e” (de, gi, le, ri, se, ne, ti, and mi) and category “B” syllables containing the vowels “o” or “u” (bo, fo, ku, mo, pu, wo, tu, and gu). The same syllables were used for both types of grammar. Examples of correct (cor) and violated (viol, in bold) sequences are given for short and long condition in both grammars.

Syntactic complexity in the brain 

an artificial grammar paradigm, brain activation elicited by local phrase structure processing and processing of more complex, hierarchical structures was investigated (Bahlmann, Schubotz, & Friederici, 2008; Friederici, Bahlmann, et al., 2006). The paradigm used in these studies employed an FSG of the (AB)n type and a PSG of the AnBn type similar to the grammars used by Fitch and Hauser (2004) (see Figure 3A). In the first of these studies (Friederici, Bahlmann, et al., 2006), category membership was marked by phonology of consonant-vowel syllables. One group of participants learned the FSG, another group learned the PSG. During the test phase in the MR scanner, each group was presented with correct and incorrect sequences of the respective grammar they had learned before. A comparison of brain activity between incorrect and correct sequences revealed an increase in activation in the FO for incorrect FSG sequences. The equivalent comparison for the PSG group revealed activation in the FO and, furthermore, activation in Broca’s area (BA 44/45) (see Figure 4). These Ungrammatical vs. grammatical sequences

Frontal operculum

Finite state grammar

Phrase structure grammar

x = -36, y = 20, z = -2

x = -36, y = 16, z = 0

Broca's area

3.09

x = -46, y = 16, z = 8

x = -46, y = 16, z = 8

Figure 4. Functional brain activation for the processing of finite state grammar (FSG) and phrase structure grammar (PSG). The Z-maps show group-averaged activation during processing of violations of both grammar types (Z > 3.09, corrected at cluster level). The contrast for incorrect vs. correct sequences in FSG (left) entail activation in the frontal operculum (FO). The PSG contrast (right) entails additional involvement of Broca’s area which is not present in the FSG. Figure modified from Friederici, Bahlmann, et al. (2006).

 Angela D. Friederici & Jens Brauer

results supported the assumption of an involvement of Broca’s area during the processing of complex, hierarchical structures. One criticism was passed on the kind of PSG of the (AB)n type used in that experiment and in Fitch and Hauser (2004). The argument was that it did not necessarily require hierarchical processes, but could be dealt with by a counting strategy what would challenge conclusions on the kind of processes subserved by the brain areas involved (de Vries, Monaghan, Knecht, & Zwitserlood, 2008; Perruchet & Rey, 2005). Therefore, a follow-up study (Bahlmann, et al., 2008) designed an extended version of the PSG grammar that employed the dependency of associated elements by a relation via voiced-unvoiced consonants. Now, in order to process a sequence of, say, A3 A2 A1 B1 B2 B3, the parser necessarily needs to build up a hierarchical relation between respective embedded elements (i.e., A1 B1 embedded within A2 B2 etc.) (see Figure 3B). Each participant learned both types of grammar (FSG and PSG, as defined above). After an implicit learning phase, they were confronted with correct and incorrect sequences and were asked to judge their grammaticality. Direct comparison between brain activation in response to grammar types showed an increase of activation for the PSG in BA 44/45. Hence, results and conclusions from the previous study (Friederici, Bahlmann, et al., 2006) were confirmed. Broca’s area is necessary to process complex hierarchical structures in contrast to local phrase structures. This functional differentiation between Broca’s area and the frontal operculum was assumed to be corroborated by structural dissociation according to cortical connectivity fingerprints of these two areas. Such an hypothesis was suggestive given MEG data (Friederici, et al., 2000) and fMRI data (Friederici, Rüschemeyer, Hahne, & Fiebach, 2003) on local phrase structure violations showing that one network might involve the FO and the anterior STG, and that another network supporting the interpretation of complex sentences might include Broca’s area (Ben-Shachar, et al., 2003; Caplan, et al., 2002; Roder, et al., 2002; Stromswold, et al., 1996) and the posterior STG, since this region was found to activate more for non-canonical object-first sentences (Bornkessel, et al., 2005; Constable, et al., 2004; Cooke, et al., 2002). In order to explore both inferior frontal areas’ (Broca and FO) connectivity patterns, DTI data from the participants of the Friederici, Bahlmann, et al. (2006) study were acquired and examined. Fiber tractography of corresponding white matter connections were analyzed for both functional activation centers. Results revealed two separate networks, one connecting the FO via the unicinate fasciculus to the anterior STG and further into the temporal cortex (local syntax), while the other connecting Broca’s area via the SLF to the posterior STG expanding further into the temporal cortex (complex syntax) (see Figure 5). Functionally, there is good empirical evidence for the local syntax network data (Brauer & Friederici, 2007; Friederici, et al., 2003; Friederici, et al., 2000). For the complex syntax network, there is broad evidence for the involvement of Broca’s area (Ben-Shachar, et al., 2003; Caplan, et al., 2002; Friederici, Fiebach, et al., 2006; Roder, et al.,

Syntactic complexity in the brain 

Structural connectivity: Tractography data (DTI)

Subject 1

Subject 3

Subject 1

Subject 3

Subject 2

Subject 4

Subject 2

Subject 4

from FO to STG via the uncinate fasciculus

from BA44/45 to STG via the superior longitudinal fasciculus

Figure 5. Tractography data for activation loci in Broca’s area and the frontal operculum (FO). Seed voxels for the tractograms were set to the grey-white-matter border closest to the individual maximum activation (black and green dots) for four representative participants from each group (FSG group, PSG group). The activation maxima were localized in FO (FSG group, orange) and in Broca’s area (PSG group, purple). For the FSG group, connection to the temporal lobe was tracked via the uncinate fasciculus. For the PSG group, the connection to the temporal lobe tracked the superior longitudinal fasciculus (SLF). Figure modified from Friederici, Bahlmann, et al. (2006).

2002; Santi & Grodzinsky, 2007; Stromswold, et al., 1996). However, the role of the posterior part of the STG for complex syntax is still in need for further clarification. This region is activated when the parser encounters ungrammatical strings of natural language (Schlesewsky & Bornkessel, 2004), when processing syntactically complex object-first compared to subject-first sentences (Bornkessel, et al., 2005; Constable, et al., 2004; Cooke, et al., 2002). But it was also observed to participate in the processing of selectional restrictions of verbs (Brauer & Friederici, 2007; Friederici, et al., 2003), as a function of verb complexity (Ben-Shachar, et al., 2003), and of verb-based argument hierarchies (Bornkessel, et al., 2005). Thus it appears that the posterior STG’s function is to subserve the integration of syntactic and verb-based syntax-relevant information during sentence comprehension.

5. Conclusion Based on available ontogenetic and phylogenetic data, one may hypothesize that the SLF plays a crucial role in the neural network supporting the processing of complex

 Angela D. Friederici & Jens Brauer

syntax. However, the functional relevance of this connection between Broca’s area and the posterior STG is yet to be determined in more detail. Ontogenetically, the temporal white matter (including the SLF) underlying language-related areas is very late in its maturation compared to central cortex white matter underlying motor or sensory functions (Pujol, et al., 2006). Even children at age 7 still show differences in myelination of the SLF compared to adults which appears to influence functional processing (Brauer, Anwander, & Friederici, 2008). Behavioral studies indicate that children only begin to process more complex structures such as passive phrases between 3.5 and 4 years (Fox & Grodzinsky, 1998). Even beyond the age of 5 years, children show problems with the processing of case-marked object-first sentences (Dittmar, et al., 2008). These observations suggest that there might be a relation between functional development and structural maturation, but this has not yet been unequivocally demonstrated and requires further correlational studies. Hitherto, no direct causal inferences can be drawn in any direction (Aslin & Schlaggar, 2006). In the macaque brain, the SLF does not project to the STG, but to the parietal cortex (Schmahmann, et al., 2007). A recent DTI study comparing fiber connections between the inferior frontal cortex and the temporal cortex in the human and non-human brain reported substantial differences between species concerning the strength of the SLF and its projections into the temporal cortex (Rilling, et al., 2008). So far, empirical evidence supports the assumption of two separate functional brain networks underlying syntactic processes. Local phrase structure building involves the frontal operculum and the anterior part of the STG, while the processing of complex hierarchical structures requires the participation of Broca’s area and very likely the posterior part of the STG. These functional networks correspond to structural connections in the brain between associated areas within each of these circuits. Relevant data from the non-human brain and from children support the view that the network responsible for complex syntax might have evolved late during phylogeny and only develops late during ontogeny. However, this concept still needs further empirical support.

References Amunts, K., Schleicher, A., Burgel, U., Mohlberg, H., Uylings, H.B.M. & Zilles, K. 1999. Broca’s region revisited: Cytoarchitecture and intersubject variability. Journal of Comparative Neurology 412(2): 319–341. Arbib, M. & Bota, M. 2003. Language evolution: Neural homologies and neuroinformatics. Neural Networks 16(9): 1237–1260. Aslin, R.N. & Schlaggar, B.L. 2006. Is myelination the precipitating neural event for language development in infants and toddlers? Neurolog 66(3): 304–305.

Syntactic complexity in the brain 

Bahlmann, J., Schubotz, R.I. & Friederici, A.D. 2008. Hierarchical sequencing engages Broca’s area. NeuroImage 42: 525–534. Ben-Shachar, M., Hendler, T., Kahn, I., Ben-Bashat, D. & Grodzinsky, Y. 2003. The neural reality of syntactic transformations: Evidence from functional magnetic resonance imaging. Psychological Science 14(5): 433–440. Bornkessel, I., Zysset, S., Friederici, A.D., von Cramon, D.Y. & Schlesewsky, M. 2005. Who did what to whom? The neural basis of argument hierarchies during language comprehension. NeuroImage 26(1): 221–233. Brauer, J., Anwander, A. & Friederici, A.D. 2008. Functional development and structural maturation of language areas in the human brain. NeuroImage 41(S1): S104. Brauer, J. & Friederici, A.D. 2007. Functional neural networks of semantic and syntactic processes in the developing brain. Journal of Cognitive Neuroscience 19(10): 1609–1623. Brauer, J., Neumann, J. & Friederici, A.D. 2008. Temporal dynamics of perisylvian activation during language processing in children and adults. NeuroImage 41(4): 1485–1493. Brodmann, K. 1909. Vergleichende Lokalisationslehre der Großhirnrinde. Leipzig: Barth. Caplan, D., Vijayan, S., Kuperberg, G., West, C., Waters, G., Greve, D. et al. 2002. Vascular responses to syntactic processing: Event-related fMRI study of relative clauses. Human Brain Mapping 15(1): 26–38. Catani, M., Jones, D.K., Donato, R. & ffytche, D.H. 2003. Occipito-temporal connections in the human brain. Brain 126(Part 9): 2093–2107. Christiansen, M.H. & Kirby, S. 2003. Language evolution: Consensus and controversies. Trends in Cognitive Sciences 7(7): 300–307. Constable, R.T., Pugh, K.R., Berroya, E., Mencl, W.E., Westerveld, M., Ni, W.J. et al. 2004. Sentence complexity and input modality effects in sentence comprehension: an fMRI study. Neuroimage 22(1): 11–21. Cooke, A., Zurif, E.B., DeVita, C., Alsop, D., Koenig, P., Detre, J. et al. 2002. Neural basis for sentence comprehension: Grammatical and short-term memory components. Human Brain Mapping 15(2): 80–94. de Vries, M.H., Monaghan, P., Knecht, S. & Zwitserlood, P. 2008. Syntactic structure and artificial grammar learning: The learnability of embedded hierarchical structures. Cognition 107(2): 763–774. Dittmar, M., Abbot-Smith, K., Lieven, E. & Tomasello, M. 2008. Young German children’s early syntactic competence: A preferential looking study. Developmental Science 11(4): 575–582. Fitch, W.T. & Hauser, M.D. 2004. Computational constraints on syntactic processing in a nonhuman primate. Science 303(5656): 377–380. Fox, D. & Grodzinsky, Y. 1998. Childrens passive - a view from the by-phrase. Linguistic Inquiry 29(2): 311–332. Frazier, L. 1987. Structure in auditory word recognition. Cognition 25(1–2): 157–187. Friederici, A.D. 1995. The time course of syntactic activation during language processing - a model based on neuropsychological and neurophysiological data. Brain & Language 50(3): 259–281. Friederici, A.D. 2002. Towards a neural basis of auditory sentence processing. Trends in Cognitive Sciences 6(2): 78–84. Friederici, A.D. 2004a. The neural basis of syntactic processes. In Cognitive Neurosciences, Vol. 3, M.S. Gazzaniga (Ed.), 789–801. Cambridge MA: The MIT Press. Friederici, A.D. 2004b. Processing local transitions versus long-distance syntactic hierarchies. Trends in Cognitive Sciences 8(6): 245–247.

 Angela D. Friederici & Jens Brauer Friederici, A.D., Bahlmann, J., Heim, S., Schubotz, R.I. & Anwander, A. 2006. The brain differentiates human and non-human grammars: Functional localization and structural connectivity. PNAS 103(7): 2458–2463. Friederici, A.D., Fiebach, C.J., Schlesewsky, M., Bornkessel, I.D. & von Cramon, D.Y. 2006. Processing linguistic complexity and grammaticality in the left frontal cortex. Cereb. Cortex 1709–1717. Friederici, A.D., Hahne, A. & Mecklinger, A. 1996. Temporal structure of syntactic parsing - early and late event-related brain potential effects. Journal of Experimental Psychology: Learning, Memory, & Cognition 22(5): 1219–1248. Friederici, A.D., Hahne, A. & von Cramon, D.Y. 1998. First-pass versus second-pass parsing processes in a wernickes and a brocas aphasic - electrophysiological evidence for a double dissociation. Brain & Language 62(3): 311–341. Friederici, A.D. & Kotz, S.A. 2003. The brain basis of syntactic processes: Functional imaging and lesion studies. Neuroimage 20(Suppl 1): S8-S17. Friederici, A.D., Pfeifer, E. & Hahne, A. 1993. Event-related brain potentials during natural speech processing - effects of semantic, morphological and syntactic violations. Cognitive Brain Research 1(3): 183–192. Friederici, A.D., Rüschemeyer, S.A., Hahne, A. & Fiebach, C.J. 2003. The role of left inferior frontal and superior temporal cortex in sentence comprehension: Localizing syntactic and semantic processes. Cerebral Cortex 13(2): 170–177. Friederici, A.D., von Cramon, D.Y. & Kotz, S.A. 1999. Language related brain potentials in patients with cortical and subcortical left hemisphere lesions. Brain 122(Part 6): 1033–1047. Friederici, A.D., Wang, Y.H., Herrmann, C.S., Maess, B. & Oertel, U. 2000. Localization of early syntactic processes in frontal and temporal cortical areas: A magnetoencephalographic study. Human Brain Mapping 11(1): 1–11. Giorgio, A., Watkins, K.E., Douaud, G., James, A.C., James, S., De Stefano, N. et al. 2008. Changes in white matter microstructure during adolescence. NeuroImage 39(1): 52–61. Grodzinsky, Y. & Friederici, A.D. 2006. Neuroimaging of syntax and syntactic processing. Current Opinion in Neurobiology 16(2): 240–246. Hagoort, P. 1993. Impairments of lexical semantic processing in aphasia - evidence from the processing of lexical ambiguities. Brain & Language 45(2): 189–232. Hahne, A., Eckstein, K. & Friederici, A.D. 2004. Brain signatures of syntactic and semantic processes during children’s language development. Journal of Cognitive Neuroscience 16(7): 1302–1318. Hahne, A. & Friederici, A.D. 1999. Electrophysiological evidence for two steps in syntactic analysis: Early automatic and late controlled processes. Journal of Cognitive Neuroscience 11(2): 194–205. Hauser, M.D., Chomsky, N. & Fitch, W.T. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298(5598): 1569–1579. Kaan, E., Harris, A., Gibson, E. & Holcomb, P. 2000. The P600 as an index of syntactic integration difficulty. Language & Cognitive Processes 15(2): 159–201. Lebel, C., Walker, L., Leemans, A., Phillips, L. & Beaulieu, C. 2008. Microstructural maturation of the human brain from childhood to adulthood. NeuroImage 40(3): 1044–1055. Matelli, M., Luppino, G. & Rizzolatti, G. 1985. Patterns of cytochrome oxidase activity in the frontal agranular cortex of the macaque monkey. Behavioural Brain Research 18(2): 125–136. Neville, H.J., Nicol, J.L., Barss, A., Forster, K.I. & Garrettdew, M.F. 1991. Syntactically based sentence processing classes: Evidence from event-related brain potentials. Journal of Cognitive Neuroscience 3: 151–165.

Syntactic complexity in the brain 

Oberecker, R. & Friederici, A.D. 2006. Syntactic event-related potential components in 24-month-olds’ sentence comprehension. Neuroreport 17(10): 1017–1021. Oberecker, R., Friedrich, M. & Friederici, A.D. 2005. Neural correlates of syntactic processing in two-year-olds. Journal of Cognitive Neuroscience 17(10): 1667–1678. Osterhout, L. & Holcomb, P.J. 1993. Event-related potentials and syntactic anomaly - Evidence of anomaly detection during the perception of continuous speech. Language & Cognitive Processes 8(4): 413–437. Osterhout, L., Holcomb, P.J. & Swinney, D.A. 1994. Brain potentials elicited by garden-path sentences: Evidence of the application of verb information during parsing. Journal of Experimental Psychology: Learning, Memory, & Cognition 20(4): 786–803. Perruchet, P. & Rey, A. 2005. Does the mastery of center-embedded linguistic structures distinguish humans from nonhuman primates? Psychonomic Bulletin & Review 12(2): 307–313. Petrides, M., Cadoret, G.V. & Mackey, S. 2005. Orofacial somatomotor responses in the macaque monkey homologue of Broca’s area. Nature 435(7046): 1235–1238. Petrides, M. & Pandya, D.N. 1994. Comparative cytoarchitectonic analysis of the human and the macaque frontal cortex. In Handbook of Neuropsychology, Vol. 9, F. Boller & J. Grafman (Eds), 17–58. Amsterdam: Elsevier. Pinker, S. & Bloom, P. 1990. Natural language and natural selection. Behavioral and Brain Sciences 13(4): 707–726. Pujol, J., Soriano-Mas, C., Ortiz, H., Sebastian-Galles, N., Losilla, J.M. & Deus, J. 2006. Myelination of language-related areas in the developing brain. Neurology 66(3): 339–343. Rilling, J.K., Glasser, M.F., Preuss, T.M., Ma, X.Y., Zhao, T.J., Hu, X.P. et al. 2008. The evolution of the arcuate fasciculus revealed with comparative DTI. Nature Neuroscience 11(4): 426–428. Roder, B., Stock, O., Bien, S., Neville, H. & Rosler, F. 2002. Speech processing activates visual cortex in congenitally blind humans. European Journal of Neuroscience 16(5): 930–936. Sanides, F. 1962. Die Architektonik des menschlichen Stirnhirns. Berlin: Springer. Santi, A. & Grodzinsky, Y. 2007. Working memory and syntax interact in Broca’s area. Neuroimage 37(1): 8–17. Schlesewsky, M. & Bornkessel, I. 2004. On incremental interpretation: Degrees of meaning accessed during sentence comprehension. Lingua 11(49–10): 1213–1234. Schmahmann, J.D., Pandya, D.N., Wang, R., Dai, G., D’Arceuil, H.E., de Crespigny, A.J. et al. 2007. Association fibre pathways of the brain: parallel observations from diffusion spectrum imaging and autoradiography. Brain 130(Part 3): 630–653. Silva-Pereyra, J., Rivera-Gaxiola, M. & Kuhl, P.K. 2005. An event-related brain potential study of sentence comprehension in preschoolers: Semantic and morphosyntactic processing. Cognitive Brain Research 23(2–3): 247–258. Stromswold, K., Caplan, D., Alpert, N. & Rauch, S. 1996. Localization of syntactic comprehension by positron emission tomography. Brain & Language 52(3): 452–473. Tomasello, M. 2007. If they’re so good at grammar, then why don’t they talk? Hints from apes’ and humans’ use of gestures. Language Learning and Development 3(2): 133–156. Ye, Z., Luo, Y.J., Friederici, A.D. & Zhou, X.L. 2006. Semantic and syntactic processing in Chinese sentence comprehension: Evidence from event-related potentials. Brain Research 1071(1): 186–196.

part iv

Biology and evolution

Neural plasticity The driving force underlying the complexity of the brain Nathan Tublitz

University of Oregon The ability of the human nervous system to process information, perform complicated, simultaneous mental and physical tasks, and express feelings and emotions is peerless. Because of its complexity, the human brain is the seminal achievement of biological evolution on our planet. This paper focuses on one aspect of brain complexity, neural plasticity, the ability of the nervous system to alter its output in response to changing stimuli. Several examples of neuroplasticity at the molecular, cellular, systems and cognitive levels are presented, all of which have physiological and behavioral consequences. The examples presented provide a basis for the premise that neural complexity arose from the need to perform complex functions. These examples also lend support for the notion that complex adaptive functions are subdivided into separate neural pathways which are oftentimes anatomically distinct.

1. Introduction “Men ought to know that from nothing else but the brain come joys, delights, laughter and sports, and sorrows, griefs, despondency, and lamentations. And by this, in an especial manner, we acquire wisdom and knowledge, and see and hear and know what are foul and what are fair, what are bad and what are good, what are sweet and what are unsavory. … And by the same organ we become mad and delirious, and fears and terrors assail us. …All these things we endure from the brain when it is not healthy. …In these ways I am of the opinion that the brain exercises the greatest power in the man.”

Hippocrates, On the Sacred Disease (4th Century B.C.)

The human brain is extraordinary. Its ability to detect and process sensory information, execute complex motor tasks, express emotions and feelings, communicate with others and generate a state of consciousness is without equal. We see and hear clearly in real time; we perform intricate motor behaviors; we feel pain, get angry, express joy; we converse freely; we think. Complex activities such as these are performed constantly by the brain and provide the underpinning for our existence as sentient beings.

 Nathan Tublitz

The total number of conscious and unconscious behaviors and actions generated by the brain has not been calculated but is likely to be astoundingly large. Even more impressive is the brain’s ability to vary the performance of every behavior depending on changes in internal or external conditions. For example, the precise muscular movements underlying running depend on the substrate; running on sand utilizes a different pattern of leg muscle activation than running on a flat track surface (Bartlett et al., 2007). Food and drink ingestion, commonly known as eating and drinking, describe ~250,000 different mouth movements (van der Bilt et al., 2006). Even the precise body position when seated in a chair is dependent on many variables including level of exhaustion and chair shape (Shenoy & Aruin, 2007). Variations in individual behaviors, known as behavioral plasticity, are not limited only to motor movements. Responses to individual sensory inputs as well as the expression of emotions and feelings are also quite variable. The exact nature of any specific motor, sensory or emotional response depends on a combination of environmental conditions and internal motivational and physiological circumstances. Arguably the best example of behavioral plasticity is human language. The foremost English language resource, the Oxford English Dictionary (2nd edition, 1989), lists over 620,000 English words. Other less conservative sources suggest there are at least 996,000 words in the English language (The Global Language Monitor, www. languagemonitor.com). The number of possible English sentences generated from this huge word base, although not infinite, is incalculably high. Much of this is due to the embedding of clauses inside other clauses, known as the recursive nature of language (Chomsky, 1965; Givon, 2008). The massive number of sentences in a given language and other forms of behavioral plasticity are possible only because of the ability of the nervous system to modify its output. Minute adjustments in motor behaviors, sensory reactions, emotional responses and conscious thought are mediated by numerous brain mechanisms. The term “neuroplasticity” is often used to characterize the neural adaptations that enable the central nervous system (CNS) to generate variations in individual behaviors. Neuroplasticity occurs at all levels of the nervous system including molecular, cellular, systems and cognitive levels. This paper presents an example of neuroplasticity at each of these levels. The premise underlying each example is that neural plasticity arose in each case due to the need to perform complex functions. Evidence is also presented for the notion that complex adaptive functions are frequently subdivided into separate neural pathways.

2. M olecular plasticity: Evolution of voltage-gated sodium channels in eukaryotes The vast diversity of life forms on our planet is thought to have evolved from one simple organism. This means that seemingly unrelated organisms such as blue-green

Neural plasticity and brain complexity 

algae, fireflies, and orangutans are distantly related and share a common ancestor. The process of gradual divergence by which more complex organisms arose from simpler ones was first described by Charles Darwin as “descent with modification” or evolution. Evolutionary change is driven by modifications at the genetic level. Random genetic alterations occasionally generate positive adaptations that, over time, lead to new, increasingly complex species (e.g., Barton, 2008). An example of a molecular level change that impacted organismal complexity is the evolution of voltage-gated sodium channels in nerve cells. Individual nerve cells have four functional regions: dendrites, which receive an input from other neurons; the cell body, which integrates all the dendritic inputs and provides other cellular functions; the axon, which connects the input and output regions of the cell; and the terminal, which sends an output signal to the next neuron (Kandel, Schwartz & Jessell, 2000). Nerve cell inputs and outputs are usually mediated by chemical neurotransmitters. To transfer the transmitter-mediated input signal from the dendrites to the terminals, a distance of 1 meter or more in some neurons, each nerve cell converts the neurotransmitter chemical message into an electrical signal which very rapidly travels down the nerve axon to the terminal (Kandel, Schwartz & Jessell, 2000). Once the electrical signal, known as an action potential, reaches the terminal, it is converted back into a transmitter-mediated chemical message that is released from the cell, crosses the synaptic cleft and is detected by the downstream neuron. The action potential of a nerve cell is generated by a rapid influx of sodium ions across the cell membrane followed closely in time by an efflux of potassium ions. Sodium and potassium each pass across the cell membrane through separate, proteinaceous pores or channels that selectively pass only one type of ion. The pores passing sodium and potassium each open and close in response to changes in voltage (Hille, 1989; 2001). When the neuron is at rest, these channels are closed. However, when the voltage changes across the cell membrane, the potassium and sodium channels open, albeit with different kinetics. Because they are activated by changes in membrane voltage, these ion-passing membrane pores are known as voltage-gated sodium and voltage-gated potassium channels, respectively (Hille, 1989; 2001). Voltage-gated sodium channels have their origin in potassium channels from prokaryotes such as bacteria (N.B., prokaryotes are organisms whose cells lack a nucleus). Bacteria typically have cellular requirements for potassium but not for sodium. Consistent with their needs, bacteria do not usually express sodium channels but do contain potassium-specific channels in their membranes to allow passage of potassium ions into and/or out of the cell (Milkman, 1994). Ranganathan (1994) suggested that the primordial prokaryotic ion channel was a potassium channel with a gate on the cytoplasmic or inside face of the channel that opens or closes it. Other researchers have suggested that other types of potassium channels with a cytoplasmic gate, such as mechanosensory or cyclic-nucleotide-gated channels,

 Nathan Tublitz

may be the ancestor of all voltage sensitive channels (Jan & Jan, 1994; Anderson & Greenberg, 2001). The general consensus, however, is that the first channels in bacteria were simple, non-voltage-gated potassium channels with a cytoplasmic gate controlling ion flow through the channel (Hille, 1989). Voltage-activated membrane channels first appeared in single celled eukaryotes such as protozoans (N.B., eukaryotes are organisms whose cells possess nuclei and other membrane bound intracellular organelles). Many protozoans express separate voltage-gated potassium and voltage-gated calcium channels (Figure 1; Hille, 1989; Anderson & Greenberg, 2001). The latter are used to bring calcium into the cell and as the charge carrier for electrical signaling purposes. There is good evidence supporting the notion that voltage-gated calcium channels evolved in primitive, singlecelled eukaryotes after the appearance of voltage-gated potassium channels and before voltage-gated sodium channels. The use of sodium as a charge carrier to change the membrane potential of a cell did not appear in evolution until the advent of multi cellularity. Sodium channels were not common until the appearance of cnidarians (N.B., an animal phylum with two cellular layers, stinging cells, and radial symmetry, e.g., hydras and jellyfish) and are not found in higher plants, protozoa and algae (Hille, 1989). Thus, voltage-gated sodium channels did not evolve until much later. These conclusions are supported by molecular data. Voltage-gated potassium channels in both prokaryotes and eukaryotes consist of multiple, independent subunits that associate together using weak chemical bonds to create the ion channel. The predominant class of voltage-activated potassium channels has four quite similar protein subunits (Figure 1). Each subunit contains 6 transmembrane spanning segments (S1–S6) plus a loop, called the P- or pore region, between segments S5 and S6 that allows the channel to be selectively permeable only to potassium. Studies have determined that the voltage sensor is localized to the S4 transmembrane region of each subunit (Figure 1; Hille, 1989; 2001). In contrast, voltage-gated calcium and sodium channels consist of a single protein with 4 very similar or homologous domains strung together linearly (Figure 1). Each domain consists of the same general architecture as that for the voltage-gated potassium subunits, i.e., 6 membrane-spanning regions (S1–S6), a P-region between S5 and S6 segments, and voltage sensitive S4 segment (Goldin, 2002). On the basis of structural and amino acid sequence data, Strong and colleagues (1993) hypothesized that the single domain, voltage-gated potassium channel subunit is likely to be the ancestral form of all voltage-gated channels. They proposed that voltage-gated calcium channels, with 4 nearly identical domains, evolved from two rounds of gene duplication during the evolution of the early prokaryotes. The first duplication produced a two-domain channel similar to domains I/III and II/IV of a voltage-gated calcium channel. Each of these two-domain channels went through another round of gene duplication to generate the first 4 domain calcium channel.

Neural plasticity and brain complexity 

Sodium Channel

I

II

III

IV

12345 P 6

12345 P 6

12345 P 6

12345 P 6

Extracellular side

Cytoplasmic side COOH

NH2

Calcium Channel

1 2345 P 6

1 2345 P 6

1 2345 P 6

1 2345 P6

COOH

NH2

Potassium Channel 1 2345 P 6

NH2

COOH

Figure 1. Schematic drawing of voltage-gated sodium, calcium and potassium channels. Sodium and calcium voltage-gated channels consist of 4 similar domains (I–IV). Each domain is composed a single polypeptide chain with 6 α-helical, transmembrane spanning regions (1–6), a pore region (P) between α-helices 5 and 6, and voltage sensor in α-helix 4 (red). The potassium channel is a single subunit containing a single repeat of the 6 α-helices. Four of these potassium subunits assemble together to form the potassium channel. NH2 and COOH refer to the amino (beginning) and carboxyl (end) terminals of the channel proteins (modified slightly from Kandel, Schwartz & Jessell, 2000).

 Nathan Tublitz

Molecular analyses of the amino acid structure of the voltage-gated sodium channel indicate that each of its four domains is more similar to its homologous domain in the voltage-gated calcium channel than to each other. These data argue in favor of the notion that voltage-gated sodium channels evolved after the double gene duplication event that created the voltage-gated calcium channel (Hille, 1989; Strong et al., 1993). These data taken together provide strong evidence that the prokaryotic potassium channel was the ancestral form of all voltage-gated ion channels. Soon after the appearance of eukaryotes, the single domain potassium channel developed voltage sensitivity followed by two rounds of gene duplication to produce the four domain, voltage-gated calcium channel. A final duplication event and alteration of ion selectivity led to the appearance of the voltage-gated sodium channel found in all invertebrate and vertebrates including humans. The advent of voltage-gated sodium channels enabled long distance, rapid signaling between cells and clearly resulted in an increase in cellular neuroplasticity and organismal complexity. The evolution of four domain, voltage-gated sodium channels in eukaryotes from single domain, non-voltage-gated potassium channels in prokaryotes is an unambiguous example of increasing biological complexity driven by evolutionary pressures.

3. C ellular neuroplasticity: Functional plasticity in mature insect neurons Structural and functional alterations at the molecular level, as illustrated by the origin of voltage-gated sodium channels discussed in the previous section, have been a primary mechanism used repeatedly throughout evolution to increase nervous system complexity. Molecular modifications, however, are not the only means to achieve the vast neural and behavioral plasticity seen in invertebrates and vertebrates, including mammals and humans. The ability of individual cells to alter their function provides another mechanism to generate complexity and behavioral plasticity. This section details an unusual example of a set of identified neurons that undergo a dramatic transformation at a specific point in their lifetime and assume a completely different identity and function. Following their birth from ectodermal cells, newly born nerve cells develop into mature adult neurons with a distinctive set of biochemical, anatomical and physiological characteristics (Kandel, Schwartz & Jessell, 2000). The combination of transmitter identity, dendritic arbor shape, axonal projection pattern, biochemical profile, and electrical properties are often sufficient to differentiate a neuron from its neighbors and identify it individually. Research on identified neurons has stimulated significant progress in cellular neurobiology. The strength of this approach lies in the ability to repeatedly analyze the same, uniquely defined neuron from different individuals of the same species. The

Neural plasticity and brain complexity 

most notable example is the squid giant axon which was used to elucidate the basic cellular properties of neurons (e.g., Hodgkin & Huxley, 1952). With few notable exceptions, invertebrate preparations have provided most of the known individually identifiable neurons because of the simplified nature of their CNS. Insects have long been a favorite model system for studies on identified cells because of their rapid generation time, ease of rearing, wide repertoire of complex behaviors, and significantly fewer neurons compared to vertebrates (~104 vs. ~1011 cells). One insect species, the tobacco hawkmoth Manduca sexta, is a particularly amenable animal model system because of the wealth of information already known about its physiology, anatomy, development, endocrinology and biochemistry of its individually identifiable neurons (e.g., Dai et al., 2007; Duve et al., 2005). The Manduca nervous system follows the general plan of other arthropods (Chapman, 1991). It consists of a cephalized brain connected to a series of segmentally iterated ganglia lying along the ventral side of the body, the latter of which are collectively known as the ventral nerve cord (Figure 2). Each ganglion in the Manduca ventral nerve cord generally contains ~1000 individual nerve cell bodies that supply neural information for a single body segment. Some cells in each ganglion are motoneurons or sensory neurons and others are interneurons which interact with motor and sensory cells. A very small percentage of ganglionic neurons are neurosecretory cells, specialized nerve cells with the electrical activity of neurons and which also act like gland cells to release neurohormonal signals into the blood (Maddrell & Nordmann, 1979). One set of identified neurosecretory cells in Manduca are the Lateral Neurosecretory Cells (LNCs; Figure 2). The LNCs are four pairs of cells in each abdominal ganglion that go through a remarkable makeover during metamorphosis from caterpillar to adult moth (Tublitz, 1993). The LNCs in caterpillars are involved in regulating cardiac activity (Prier, Hwa & Tublitz, 1994). They release a set of hormones called Cardioacceleratory Peptides at specific times during larval life (Tublitz et al., 1991). For example, the CAPs are released into the blood to control heart rate during the last larval stage when the caterpillar is preparing to enter metamorphosis (Tublitz et al., 1992). During the metamorphic transition from caterpillar to adult moth, the LNCs stop making the CAPs and begin to produce a different neurotransmitter, bursicon, a peptide hormone involved in tanning the skin or cuticle of the adult moth (Loi & Tublitz, 1993; Tublitz & Loi, 1993). The transmitter switch from CAPs to bursicon is triggered by the insect steroid hormone 20-hydroxyecdysone, which mediates the down-regulation of CAP production and up-regulation of bursicon expression (Tublitz, 1993). The LNCs alter other properties during metamorphosis in addition to changing its transmitter profile. The dendritic arbors of the LNCs expand their penetration of the abdominal ganglion by nearly three fold during the transformation from larva to adult (Figure 3; McGraw et al., 1998). Physiological properties of the LNCs also

 Nathan Tublitz

Larva

BRAIN

SEG

T1–T3

A1

Adult

BRAIN

SEG

T1

T2–A2

A2–A6 A3–A5

A7–A9

A6–A8

Figure 2. Schematic drawing of the Manduca nervous system in larvae and adults highlighting the Lateral Neurosecretory Cells (LNCs; filled circles). The Manduca nervous system consists of a cephalized brain plus a nerve cord with individual or fused ganglia. The nerve cord contains one subesophageal ganglion (SEG), 3 thoracic ganglia (T1–T3) and 8–9 abdominal ganglia (A1–A9) which may or may not be partially fused depending on the developmental stage.

Neural plasticity and brain complexity 

change at this time. Two electrical measures, action potential threshold and input resistance, each decline significantly in the LNCs during metamorphosis (Tublitz & Prier, unpublished data). Like the transmitter switch, these other changes are triggered by 20-hydroxyecdysone. These data demonstrate that the LNCs alter their physiology, biochemistry and morphology in response to steroid hormone exposure. The changes at the cellular level exhibited by the LNCs during metamorphosis underlie a major switch in function. As described above, the primary purpose of the LNCs in larvae is to regulate heart rate via a local release of the CAPs in the neighborhood of the heart (Tublitz, 1993). In contrast, the same LNCs perform a very different

anterior

Larval

Adult

12 *

Arborization (%)

10 8 6 4 2 0

Larval

Adult

Figure 3. A comparison of the total extent of arborization in larval and adult lateral neuro secretory cells (LNCs) in the tobacco hawkmoth Manduca sexta. (A) Camera lucida drawing of the central processes of an individual LNC taken from a larva (left) and an adult (right). (B) Extent of arborization of larval and adult LNCs expressed as a percentage of the total ganglionic volume containing dendritic and axonal processes. Values are mean + S.E.M (N=5 for each data set). *The mean extent of arborization in the adult was significantly greater than that in the larva, using a one-tailed Mann-Whitney U-test, P<0.005 (taken from McGraw et al., 1998).

 Nathan Tublitz

function in adult moths, releasing the neurohormone bursicon into the blood to induce tanning of the cuticle. Thus, the LNCs act as local cardiomodulatory neurons in larvae yet serve a neurohormonal function in adults. The plasticity exhibited by the LNCs is part of a major overhaul of the Manduca central nervous system during metamorphosis. Some neurons die, others arise de novo, and still others undergo a respecification of targets and/or function (e.g., Levine, 1984; Dulcis & Levine, 2004). The LNCs fall into this latter category, changing their function from local modulator in caterpillars to neurohormonal cells in adult moths. The functional alterations seen in many Manduca neurons during metamorphosis correlate well with the massive changes in anatomy and behavior during this period. As the animal goes from the larva to pupa and finally to the adult, it undergoes a massive conversion from a feeding and crawling animal to one that flies and mates. The complexity of the anatomical and behavioral transition in Manduca is mirrored by an equally complex neural reorganization (Truman & Riddiford, 2007). The LNCs are arguably the best examples of neuronal plasticity at the cellular level in the animal kingdom, however they are not the only neurons to exhibit significant alterations in function. Functional changes by nerve cells have been described in many organisms from invertebrates to mammals (e.g., Glanzman, 2006; Kampa et al., 2007; Neves et al., 2008; Nikitin, 2007). Sensory cortex cells that have lost their sensory inputs due to lesions or injuries replace those inputs with other sensory signals, usually of the same sensory modality (e.g., Kral & Eggermont, 2007). Motor neurons innervate another muscle when their original target muscles are removed (e.g., Purves & Hadley, 1985). It is not known what percentage of neurons alter their function but the number of studies detailing cellular plasticity has certainly been on the rise. It is likely that this type of cellular neuroplasticity underlies the variation in the production of complex behaviors in many organisms.

4. Systems neuroplasticity: Body patterning behavior in cuttlefish Changes at the molecular and cellular levels explain many neurally mediated functions including sensory processing, sensorimotor integration, learning and memory, and motor behaviors. Some cases of behavioral plasticity, however, depend on groups of neurons working together as an ensemble. One example of neuroplasticity at the systems level is the neuroregulation of body patterning behavior in cephalopods molluscs, a taxonomic group that includes octopus, squid, cuttlefish and nautilus (Hanlon & Messenger, 1988). The unique combination of properties found in this group of organisms make them excellent models for studies on the neural control of behavioral plasticity at the systems level.

Neural plasticity and brain complexity 

Of the many fascinating behaviors in cephalopods, perhaps the most remarkable is their ability to rapidly produce highly detailed coloration patterns extending across their entire body. Although body patterning behavior is exhibited by all cephalopods except for the shelled nautiloids, cuttlefish are generally thought to display the largest and most complex repertoire of patterns (Hanlon, 1982; Hanlon & Messenger, 1988; Holmes, 1940; Packard & Hochberg, 1977). Most studies on body patterning behavior have been performed on one species, the European cuttlefish Sepia officinalis, which like other cephalopods, are capable of adjusting their body coloration to match numerous different substrates, including many that are visually complex (Figure 4). Sepia also display specific body patterns in response to the appearance of predators, prey and conspecifics (e.g., during courtship) as well as to local environmental disturbances. Many body patterns are stunningly dynamic; for example Sepia display a set of several dark and light transverse bands across the body that move anteriorly at 5–10 Hz, commonly known as the ‘passing cloud’ display (Packard & Hochberg, 1977). The complexity of these displays reflects a CNS origin, and most are visually mediated because blinded cephalopods exhibit significantly fewer body patterns (Sanders & Young, 1974). Cephalopod body patterns are generated by a suite of chromatic elements, including iridiphores, leucophores and chromatophores, the latter of which are responsible for the amazing ability of cuttlefish and other cephalopods to generate intricate displays in less than a second, much faster than any other species in the animal kingdom (Messenger, 2001). The ability of cuttlefish and other unshelled cephalopods to generate complex patterns so quickly is due to the unique structure of their chromatophore system. A cephalopod chromatophore is a true multicellular organ (Figure 5); at its core is the chromatophore cell, a pigment-containing cell with a highly elastic plasmalemma. Attached to and radiating from each chromatophore cell are 6–20 striated muscles cells, the chromatophore muscles, which emanate from the chromatophore cell like the spokes of a bicycle wheel (Figure 5; Cloney & Florey, 1968). Contraction or relaxation of the chromatophore muscles results in expansion or retraction, respectively, of the chromatophore cell. Because chromatophore muscles produce graded contractions, many intermediate expansion states of the chromatophore cell are possible. Individual chromatophore cells also exhibit dynamic responses, e.g., “flickering” behavior produced by rapid mini-contraction/relaxation cycles of the chromatophore muscles. Ultimate control of body patterning in unshelled cephalopods lies within the CNS since most if not all chromatophore muscles are innervated by motoneurons (Reed, 1995). Like the body patterning behavior it mediates, CNS control of chromatophore activity is elaborate. Invertebrate striated muscles, unlike their vertebrate counterparts, are innervated by multiple types of motoneurons. Sepia chromatophore muscles receive both direct excitatory and inhibitory motor input (Loi & Tublitz, 2000). There

 Nathan Tublitz

Figure 4. Photograph of a juvenile European cuttlefish Sepia officinalis producing a camouflage pattern while resting on top of pebbles.

are two different types of excitatory motoneurons, one which causes a fast contraction of the chromatophore muscles and the other which induces a slower muscular excitation. These “fast” and “slow” motoneurons are distinguished by their neurotransmitters. The fast motoneurons use the amino acid glutamate and the slow motoneurons release multiple peptides all within the FMRFamide peptide family (Figure 6; Loi & Tublitz, 1997; 2000). The third input to the chromatophore muscles is inhibitory, mediated by the classic biogenic amine neurotransmitter serotonin (Messenger, 2001). It is the interaction between these three different neural inputs that underlies the complex responses of individual chromatophore organs (Messenger, 2001). Mature adult Sepia each contain approximately 400,000 to 1 million chromatophore organs and approximately 200,000 chromatophore motoneurons (Hanlon & Messenger, 1988; Messenger, 2001). These figures suggest that each chromatophore motoneuron controls on average about 2–5 individual chromatophore organs and innervates 12–100 chromatophore muscles. Although there is some convergence of neural information from motoneuron to chromatophore organ, the number of individual motoneurons is more than sufficient to produce the complex body patterning seen in these amazing organisms. More information about the control of body patterning behavior has been discerned by the location and distribution of the chromatophore motoneurons within the Sepia CNS. The vast majority of chromatophore motoneurons have been localized to the anterior and posterior chromatophore lobes of the Sepia brain (Gaston & Tublitz, 2004). In general, the chromatophore motoneurons innervating the tentacles

Neural plasticity and brain complexity  Pigment granules

Nucleus

Cytoelastic sacculus

Primary Infoldings and Pouches

Glial cell Nerve axon

Mitochondria Muscle cell

Radial muscle fibers

Figure 5. Diagram of the ultrastructure of a retracted cephalopod chromatophore organ. The sheath cells covering the chromatophore and the muscle fibers are not shown (slightly modified from Cloney & Florey, 1968).

are found in the anterior chromatophore lobe while the posterior chromatophore lobe houses the motoneurons controlling chromatophores on the rest of the body and the fin (Boycott, 1961; Gaston & Tublitz, 2004 & 2006). Recent brain localization studies on cuttlefish fin chromatophore motoneurons suggest the presence of a somatotopic map in the posterior chromatophore lobe (Gaston & Tublitz, 2006). Somatotopy, the spatial mapping of peripheral body regions onto specific CNS regions, is a well known neural concept in vertebrates including mammals and humans (Kandel, Schwartz & Jessell, 2000). Classic examples of somatotopic mapping include the human sensory and motor cortex and retinal projections to the mammalian lateral geniculate nucleus. Although the cuttlefish data are preliminary, chromatophore motoneuron somatotopy may be present in several brain regions (Gaston & Tublitz, 2006; Tublitz, Loi & Gaston, 2006). There are several reported cases of somatotopy in invertebrate optic lobes, however this report, if confirmed, would be the first example of central somatotopic mapping in an invertebrate brain. Given that the cephalopod CNS is the most complicated and largest among the invertebrates in terms of volume and cell number (Young, 1971), it is therefore not surprising that principle of somatotopic mapping would first arise in these unshelled organisms, which require stealth, camouflage and rapid movement to avoid predation. The size and intricacy of the cephalopod CNS provide a level of complexity and computational power previously unknown in the invertebrates. Although much is still to be understood about the neural control of body patterning behavior, it remains the quintessential example of systems level neuroplasticity in invertebrates.

 Nathan Tublitz

Glutamate (10–3M) 5 min

Retracted

Expanded FMRFamide (10–6M) 5 min

Retracted

Expanded Figure 6. The effect of glutamate (top panel) and the neuropeptide FMRFamide (bottom panel) on an individual in vitro chromatophore. A piece of skin from the fin of the cuttlefish Sepia officinalis was removed, pinned to a dish and immersed in sea water. A photooptical system (Loi & Tublitz, 2000) was used to measure the expansion of an individual chromatophore. The bar above each trace indicates the period of transmitter application. Note that glutamate caused immediate expansion of the chromatophore which lasted only for the duration of its application. In contrast, the initial effect of FMRFamide was delayed and the duration of its effect was prolonged for many minutes after FRMFamide removal.

5. C ognitive neuroplasticity: The neural basis of laughter and humor in humans This paper has shown to this point how plasticity at the molecular, cellular and systems levels contributes to neural complexity. Plasticity is also a key feature of the most complicated of human brain activities such as language, learning and memory, perception, thought, and planned action. Each of these cognitive activities involves larger, interconnected networks of neurons. This section focuses on the cognitive processes and brain pathways involved in an intriguing cognitive function, humor, with a specific emphasis on laughter. Humor is a universal aspect of the human experience (Apte, 1985). It occurs in all cultures and nearly all individuals throughout the world (Lefcourt, 2001). The Oxford English Dictionary (www.oed.com) defines humor as “that quality of action, speech or writing which excites amusement; the faculty of perceiving what is ludicrous or amusing; or of expressing it in speech, writing or other actions”. It is evident that humor is a

Neural plasticity and brain complexity 

broad term encompassing anything people do or say that is perceived as amusing and which tends to make people laugh. Humor also includes the mental processes that go into creating and perceiving amusing stimuli and the physical responses involved in enjoying humor. Like all mental and physical processes, humor is a complex series of actions taken by the brain. From detecting a humorous stimulus, to its processing and understanding, and finally to producing a response such as laughter, the brain is involved in every step. It is this final step, laughter and its underlying neural processes, that is the focus of this section. A frequent behavioral response to a humorous stimulus is laughter. Charles Darwin noted in his 1872 book The Expression of the Emotions in Man and Animals that laughter is a mechanism to express a specific emotional reaction to others. Laughter is a distinctive, stereotyped pattern of vocalization that is easily recognizable and quite unmistakable (Provine & Yong, 1991). Although what is funny varies greatly, the sounds of laughter are indistinguishable across cultures. Developmentally, laughter is one of the first social vocalizations after crying emitted by human infants (McGhee, 1979). Infants first laugh in response to the behavior of others at about four months old. Cases of gelastic or laughter producing epilepsy in newborns indicate that the brain mechanisms for laughter are already present at birth (Sher & Brown, 1976). The innateness of laughter is best demonstrated in those born deaf and blind whom are reported to laugh appropriately without ever having perceived the laughter of others (Provine, 2000). Laughter is a complex physiological response, characterized by a combination of loud oral noises, repetitive diaphragm contractions, open mouth and grimaces caused by facial muscle contractions and flushing of the skin (Ruch & Ekman, 2001). Laughter is also accompanied by a general physiological arousal including increased heart rate, production of tears, loss of strength in the extremities and flailing body movements (Cacioppo et al., 2000). All of these physiological activities are precisely controlled by the brain. Studies of patients with brain lesions have identified two distinct functional pathways in the brain that produce smiling and laughter. One pathway produces involuntary, spontaneous and emotional laughter known as genuine laughter. The second pathway mediates forced laughter, which is voluntary and unemotional. The functional separation of these pathways has been demonstrated in several different types of studies. Some stroke patients who are unable to voluntarily move their facial muscles (i.e., volitional facial paresis), are nonetheless capable of genuine laughter. They are able smile and laugh normally in response to a humorous stimulus (Wild et al., 2003). In contrast, some patients with subcortical brain lesions in the basal ganglia (Figure 7) are not able to exhibit spontaneous facial expressions of emotion when they feel amused but are able to smile on command (Wild et al., 2003).

 Nathan Tublitz

Supplementary Motor Area

Cingulate Cortex Parietal Lobe

Frontal Lobe Prefrontal Cortex Basal Ganglia Nucleus Accumbens

Cerebellum

Pons

Ventral Tegmental Area

Figure 7. Brain regions involved in cognitive and motor regulation of laughter (modified from Martin, 2007).

Additional evidence for separate pathways for voluntary and involuntary pathways comes from studies using positron emission tomography (PET), a brain imaging technique that measures changes in regional cerebral blood flow. Iwase and colleagues (2002) used PET to test responses of healthy individuals to humorous or non-humorous videos. Involuntary smiling specifically activated cortical areas such as the left anterior temporal cortex and bilateral occipitotemporal cortices involved in visual processing and integration (Figure 7). Limbic system areas involved in emotional reward were also activated. In contrast, voluntary, non-emotional smiling was correlated with greater activity in those areas of the frontal cortex involved in voluntary facial movement such as primary and supplementary motor areas (Figure 7). Data from electrical brain stimulation experiments lend further support to the working hypothesis that there are separate brain pathways for voluntary and involuntary laughter. One study described a 16-year-old female patient with epilepsy who, when her supplementary motor cortex was stimulated electrically, consistently laughed even though there was no detectable visual or auditory humorous stimulus (Fried et al., 1998). Her laughter was accompanied by the subjective feelings usually associated with humor such as amusement and mirth. Even more interesting was that every time she laughed due to electrical stimulation, she ascribed her laughter to a specific stimulus in the room. For example, once she claimed to be laughing because of the humorous

Neural plasticity and brain complexity 

nature of a horse photo on the wall. It must be noted that this patient’s bouts of epilepsy were never accompanied by gelastic laughter. Based on data from stroke victims, PET and fMRI brain scans, and electrical brain stimulation, it is becoming increasingly clear that the brain has two pathways for laughter that are functionally and anatomically distinct. It is reasonable to postulate that each pathway arose independently to perform its distinctive function and was maintained because it provided a selective advantage to the individual. Although the precise mechanisms and details of the networks involved in humor detection and laughter production remain to be elucidated, the presence of two separate two neural circuits for laughter provides a concrete example of the development of neural complexity at the cognitive level.

6. Conclusions A fundamental tenet of modern evolutionary thought is that complex structures and life forms arose from simpler ones. Biology is rife with microscopic and macroscopic examples of this principle: for example, the origin of energy producing cellular organelles (i.e., chloroplasts and mitochondria) from free-living aerobic bacteria; the advent of multicellularity from single-celled organisms; the appearance of bird feathers from reptilian body scales; the development of the vertebrate camera eye from simple invertebrate photoreceptors; and of course, the evolution of modern day humans from our simian ancestors. The prevailing theme in each case is that these new designs arose from structures already in existence. The basic mechanism underlying this theme is that a structure originally intended to fulfill one role is slowly changed through gradual modification to become adaptive for a different role. It is by this mechanism that simple structures evolve into more complex ones. The principle of gradual adaptation is the bedrock of modern evolution. Gradual adaptation is the force underlying evolutionary change not only in body structure but also in the form and function of the nervous system. From the simple nerve nets of cnidarians (e.g., sponges, jellyfish and corals) to the complicated interactions of the central and peripheral nervous systems in insects, to the highly cephalized brain of cephalopods with functionally discrete regions, and finally to the remarkable human brain, nervous systems have developed from the simple to the highly complex over evolutionary time. During the course of evolution, the nervous system of animals has increased in absolute size and in its number of neurons. New functions have also arisen, including the ability to alter one’s own behavior in response to changing environmental or internal conditions. Known as behavioral plasticity, this function is an essential adaptation without which individual organisms would neither survive to maturity nor reproduce.

 Nathan Tublitz

All organisms from the simplest unicellular bacteria to humans are endowed with this capacity to a lesser or greater extent. Responsibility for generating variations in behavioral output lies exclusively with the nervous system and its remarkable plasticity. This paper has presented several molecular, cellular, systems and cognitive examples of neuroplasticity underlying behavioral plasticity. Neuroplasticity arose in each case because of the necessity to perform a complex novel function. At the molecular level, voltage-gated sodium channels appeared in multicellular animals in order to coordinate responses among other cells in the organism. The need to re-specify cellular function in the nervous system of organisms with relatively few neurons is the likely reason behind the major biochemical, physiological and anatomical changes exhibited by the Manduca LNCs. The intricate, systems level regulation of cephalopod chromatophores arose because of the necessity to generate precise camouflage patterns and avoid prey. At the cognitive level, the two types of laughter, voluntary and involuntary, evolved to serve different functions by way of separate neural pathways. New functions in each instance resulted from the ability of the nervous system to adapt and adjust to changing environmental and internal conditions. In some cases new pathways were developed to accommodate these changes. Neuroplasticity at the molecular, cellular, systems and cognitive levels is an essential component for behavioral flexibility and complexity in all animals. Human language, a seminal function of the human brain, requires the activation and integration of at least 3 separate neural processes. A simple conversation between speaker and listener is successful only if (1) the incoming auditory stimulus from the speaker (i.e., spoken words) is detected by the listener’s auditory system (sensory detection); (2) the stimulus is decoded and understood by the listener (language processing); and, (3) the listener replies with a an appropriate verbal response (motor output). Each of these three steps (sensory detection, language processing and motor output) has a very high degree of variability due to vast number of possible inputs (i.e., words and sentences) and a large number of possible motor-mediated verbal replies. The language pathway thus represents a quintessential example of behavioral plasticity. Although a neuromechanistic understanding of language detection, integration and production remains elusive, it is highly probable that language complexity has its genesis in the myriad of neural mechanisms underlying brain plasticity.

Acknowledgements I thank Dr. Michelle Gaston, Dr. Poh Kheng Loi, and Mr. Kevin Prior for contributing to the some of the work presented in this paper. I also thank Ms. Alexandra Lewis for commenting on a draft of this manuscript. The work in this paper was supported by the National Institutes of Health, the National Science Foundation, and the Medical Research Foundation of Oregon.

Neural plasticity and brain complexity 

References Anderson, P.A.V. & Greenberg, R.M. 2001. Phylogeny of ion channels: clues to structure and function. Comparative Biochemistry & Physiology 129B: 17–28. Apte, M.L. 1985. Humor and Laughter: An Anthropological Approach. Ithaca NY: Cornell University Press. Bartlett, R., Wheat, J. & Robins M. 2007. Is movement variability important for sports biomechanists? Sports Biomechanics 6: 224–243. Barton, N.H. 2008. The effect of a barrier to gene flow on patterns of geographic variation. Genetic Research 90: 139–149. Boycott, B.B. 1961. The functional organization of the brain of the cuttlefish Sepia officinalis. Proceedings of the Royal Society of London B 153: 503–534. Cacioppo, J.T., Bernston, G.G., Larsen, J.T., Poehlmann, K.M. & Ito, T.A. 2000. The psychophysiology of emotion. In Handbook of emotions. M. Lewis & J.M. Haviland-Jones (Eds), 2nd Edn, 173–191. New York: Guilford. Chapman, R.F. 1991. The Insects: Structure and Function. Cambridge: Cambridge University Press. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Cloney, R.A. & Florey, E. 1968. Ultrastructure of cephalopod chromatophore organs. Zeitschrift fuer Zellforschung und Mikroskopische Anatomie 89: 250–280. Dai, L., Zitnan, D. & Adams, M.E. 2007. Strategic expression of ion transport peptide gene products in central and peripheral neurons of insects. Journal of Comparative Neurology 500(2): 353–367. Dulcis, D. & Levine, R.B. 2004. Remodeling of a larval skeletal muscle motoneuron to drive the posterior cardiac pacemaker in the adult moth, Manduca sexta. Journal of Comparative Neurology 478(2): 126–142 Duve, H., Pell, K., Hines, E., East. P. & Thorpe, A. 2005. Neuropeptide co-localisation in the lepidopteran frontal ganglion studied by electron-microscopic gold-labeling immunocytochemistry. Cell and Tissue Research 320(1): 187–199. Fried, I., Wilson, C.L., MacDonald, K.A. & Behnke, E.J. (1998). Electric current stimulates laughter. Nature 39: 650–652. Gaston, M. & Tublitz, N.J. 2004. Peripheral innervation patterns and central distribution of fin chromatophore motoneurons in the cuttlefish Sepia officinalis. Journal of Experimental Biology 207: 3089–3098. Gaston, M. & Tublitz, N.J. 2006. Somatotopic innervation of anterior fin chromatophore in the European cuttlefish, Sepia officinalis. Invertebrate Neuroscience 6: 81–93. Givón, T. 2008. The Genesis of Syntactic Complexity, Introduction. Amsterdam: J. Benjamins. Glanzman, D.L. 2006. The cellular mechanisms of learning in Aplysia: of blind men and elephants. Biological Bulletin 210(3): 271–279. Goldin, A.L. 2002. Evolution of voltage-gated Na channels. Journal of Experimental Biology 205: 575–584. Hanlon, R.T. 1982. The functional organization of chromatophores and iridescent cells in the body patterning of Loligo plei (Cephalopoda:Myopsida). Malacologia 23: 89–119. Hanlon, R.T. & Messenger, J.B. 1988. Adaptive coloration in young cuttlefish (Sepia officinalis): The morphology and development of body patterns and their relation to behavior. Philosophical Transactions of the Royal Society of London B 320: 437–487. Hille, B. 1989. Ionic channels: evolutionary origins and modern roles. Quarterly Journal of Experimental Physiology 74: 785–804.

 Nathan Tublitz Hille, B. 2001. Ion Channels of Excitable Membranes. 3rd Edn. Sunderland, MA: Sinauer Associates, Inc. Hodgkin, A.L. & Huxley, A.F. 1952. Currents carried by sodium and potassium ions through the membrane of the giant axon of Loligo. Journal of Physiology (London) 116: 449–472. Holmes, W. 1940. The color changes and color patterns of Sepia officinalis L. Proceedings of the Zoological Society of London 110: 17–35. Iwase, M., Ouchi, Y., Okada, H., Yokoyama, C., Nobezawa, S. & Yoshikawa, E. 2002. Neural substrates of human facial expression of pleasant emotion induced by comic films: A PET study. NeuroImage 17: 758–768. Jan, L.Y. & Jan, Y.N. 1994. Potassium channels and their evolving gates. Nature 371: 119–122. Kampa, B.M., Letzkus, J.J. & Stuart, G.J. 2007. Dendritic mechanisms controlling spike-timingdependent synaptic plasticity. Trends in Neuroscience 30(9): 456–463. Kandel, E.R., Schwartz, J.H. & Jessell, T.M. 2000. Principles of Neural Science. 4th Edn, New York NY: McGraw Hill. Kral, A. & Eggermont, J.J. 2007. What’s to lose and what’s to learn: development under auditory deprivation, cochlear implants and limits of cortical plasticity. Brain Research Reviews 56(1): 259–269. Lefcourt, H.M. 2001. Humor: The Psychology of Living Buoyantly. New York: Kluwer Academic. Levine, R.B. 1984. Changes in neuronal circuits during insect metamorphosis. Journal of Experimental Biology 112: 27–44. Loi, P.K. & Tublitz, N.J. 1993. Steroid regulation of transmitter phenotype in individual insect peptidergic neurons. I. The commitment peak of 20-OH ecdysone triggers the loss of CAP2 expression. Journal of Experimental Biology 181: 175–194. Loi, P.K. & Tublitz, N.J. 1997. Molecular analysis of FMRFamide and FMRFamide related peptides (FaRPs) in the cuttlefish Sepia officinalis. Journal of Experimental Biology 200: 1483–1489. Loi, P.K. & Tublitz, N.J. 2000. Neural regulation of cuttlefish chromatophore function by glutamate and FMRFamide. Journal of Comparative Neurology 420: 499–511 Maddrell, S.H.P. & Nordmann, J.J. 1979. Neurosecretion. Glasgow, Scotland: Blackie & Son Limited. Martin, R.A. 2007. The Psychology of Humor: An Integrative Approach. California: Elsevier. McGhee, P.E. 1979. Humor: Its Origin and Development. San Francisco: W.H. Freeman. Messenger, J.B. 2001. Cephalopod chromatophores: neurobiology and natural history. Biological Reviews of the Cambridge Philosophical Society 76: 473–528. McGraw, H.F., Prier, K.R.S., Wiley, J.C. & Tublitz, N.J. 1998. Steroid-regulated morphological plasticity in a set of identified peptidergic neurons in the moth Manduca sexta. Journal of Experimental Biology 201: 2981–2992. Milkman, R. 1994. An Escherichia coli homologue of eukaryotic potassium channel proteins. Proceedings of the National Academy of Sciences USA 91: 3510–3514. Neves, G., Cooke, S.F. & Bliss, T.V. 2008. Synaptic plasticity, memory and the hippocampus: a neural network approach to causality. Nature Reviews of Neuroscience 9(1): 65–75. Nikitin, V.P. 2007. A new mechanism of synapse-specific neuronal plasticity. Neuroscience of Behavior & Physiology 37(6): 559–370. Packard. A. & Hochberg, F.G. 1977. Skin patterning in Octopus and other genera. Symposia of the Zoological Society of London 38: 191–231. Prier, K.R., Hwa, O. & Tublitz, N.J. 1994. Modulating a modulator: Biogenic amines at subthreshold levels potentiate peptide-mediated cardioexcitation in an insect heart. Journal of Experimental Biology 197: 377–392.

Neural plasticity and brain complexity 

Provine, R.R. 2000. Laughter: A Scientific Investigation. New York: Penguin. Provine, R.R. & Yong, Y.L. 1991. Laughter: A stereotyped human vocalization. Ethology 89(2): 115–124. Purves, D. & Hadley, R.D. 1985. Changes in the dendritic branching of adult mammalian neurones revealed by repeated imaging in situ. Nature 315: 404–406. Ranganathan, R. 1994. Evolutionary origins of ion channels. Proceedings of the National Academy of Sciences USA 91: 3484–3486. Reed, C.M. 1995. The ultrastructure and innervation of muscles controlling chromatophore expansion in the squid, Loligo vulgaris. Cell and Tissue Research 282: 503–512. Ruch, W. & Ekman, P. 2001. The expressive pattern of laughter. In Emotion, Qualia and Consciousness, A. Kasniak (Ed.), 426–443. Tokyo: World Scientific. Sanders, G.D. & Young, J.Z. 1974. Reappearance of specific colour patterns after nerve regeneration in Octopus. Proceedings of the Royal Society of London B 186: 1–11. Shenoy, S. & Aruin, A.S. 2007. Effect of chair design on feed-forward postural control in sitting. Motor Control 11: 309–21. Sher, P.K. & Brown, S.B. 1976. Gelastic epilepsy: Onset in neonatal period. American Journal of Diseases in Childhood 130: 1126–1131. Strong, M., Chandy, K.G. & Gutman, G.A. 1993. Molecular evolution of voltage-sensitive ion channel genes: on the origins of electrical excitability. Molecular Biological Evolution 10: 221–242. Truman, J.W. & Riddiford, L.M. 2007. The morphostatic actions of juvenile hormone. Insect Biochemistry and Molecular Biology 37: 761–770. Tublitz, N.J. 1993. Steroid-induced plasticity in insect peptidergic neurons. Comparative Biochemistry & Physiology 105C: 147–154. Tublitz, N.J., Allen, A.T., Cheung, C.C., Edwards, K.K., Kimble, D.P., Loi, P.K. & Sylwester, A.W. 1992. Insect cardioactive peptides: regulation of hindgut activity by cardioacceleratory peptide 2 (CAP2) during wandering behaviour in Manduca sexta larvae. Journal of Experimental Biology 165: 241–264. Tublitz, N.J., Brink, D., Broadie, K.S., Loi, P.K. & Sylwester, A.W. 1991. From behavior to molecules: An integrated approach to the study of neuropeptides. Trends in Neuroscience 14: 254–259. Tublitz, N.J. & Loi, P.K. 1993. Steroid regulation of transmitter phenotype in individual insect peptidergic neurons. II. The prepupal peak of 20-OH ecdysone directly induces bursicon expression. Journal of Experimental Biology 181: 195–213. Tublitz, N.J., Loi, P.K. & Gaston, M.R. 2006. Neural regulation of complex behavior: body patterning in cephalopod molluscs. Integrative & Comparative Biology 46: 880–889. van der Bilt, A., Engelen, L., Pereira, L.J., van der Glas, H.W. & Abbink, J.H. 2006. Oral physiology and mastication. Physiology & Behavior 89: 22–27. Wild, B., Rodden, F.A., Grodd, W. & Ruch, W. 2003. Neural correlates of laughter and humour. Brain 126: 212–2138. Young, J.Z. 1971. The Anatomy of the Nervous System of Octopus vulgaris. Oxford: Clarendon Press.

Recursion Core of complexity or artifact of analysis? Derek Bickerton

University of Hawaii

1. Introduction Several years ago, there appeared in the prestigious journal Science, which does not normally pay much attention to language, an article co-authored by Marc Hauser, Noam Chomsky and Tecumseh Fitch somewhat portentously entitled “The Faculty of Language: What Is It, Who Has It, and How Did It Evolve?” (Hauser, Chomsky & Fitch 2002, henceforth HCF) The article was placed in that section of the journal titled “Science’s Compass,” and it was indeed designed to give directions to us poor benighted folks who (unlike the authors of the article) had actually been laboring in the quagmire of language evolution studies for a number of years. The paper sought to derive the computational component of language (that is, what gives language its complexity) from a single process: recursion. The paper divided language into two parts: FLN (narrow faculty of language) and FLB (broad faculty of language) • FLB = all the parts of language either not unique to humans or human but not uniquely involved in language • FLN = all the parts of of language uniquely human and uniquely linguistic The working hypothesis:of the paper was that the sole content of FLN is recursion. Recursion, in turn, might well prove to be the exaptation of a faculty found in other species but used by them for non-linguistic purposes. Number, navigation, and social interaction were some of the functions suggested.

2. Some background In order to understand where HCF is coming from, some background information is necessary. Chomsky had for years avoided committing himself on language evolution. During the 1990s he saw the field expanding, making him irrelevant. The logic of minimalism

 Derek Bickerton

forced him to become a player, but he needed leverage from biology to achieve a commanding position via the pages of Science. Prior to 2002, he and Hauser had been on opposite sides of most issues. Hauser believed that language was on a continuum with animal communication and had emerged through natural selection. Chomsky believed language was totally distinct from animal communication and did not believe that language had been specifically selected for. HCF represented a strategic compromise. C yielded to H on most aspects of language but preserved what was most vital to him: a unique central process for syntax, one that had not been specifically selected for as a component of language, thus preserving intact his claim of uniqueness and independence from natural selection over a more limited domain.

3. Defining recursion But what exactly is recursion? More than one commentator has expressed concern over the vagueness of HCF with regard to definitions. The following are the clearest indications the paper offers: “..[Recursion] provid[es] the capacity to generate an infinite range of expressions from a finite set of elements…” “All approaches agree that a core property of FLN is recursion, attributed to narrow syntax in the conception just outlined. FLN takes a finite set of elements and yields a potentially infinite array of discrete expressions.” This differs from the usual definitions of recursion within a linguistic sphere of reference. Three typical (and very recent) examples follow. “In fact, we can embed one sentence inside another again and again without limit, if we are so inclined! This property of syntactic rules is known as recursion.” (Phillips, n.d.) “In linguistics, this term refers to the fact that a sentence or phrase can contain (embed) another sentence or phrase — much like a box within a box, or a picture of someone holding a picture. Common recursive structures include (1) subordinate clauses; e.g., He said that she left, where she left is itself a sentence; (2) relative clauses; e.g., She’s the one who took the book.” (Levy 2007) “While iteration simply involves repeating an action or object an arbitrary number of times, recursion involves embedding the action or object within another instance of itself.” (Parker 2007) A feature common to all these definitions (and many others in the literature) is the insertion of something within something else of the same kind. The resulting constructions are, of course, a major source of complexity in syntax.

Recursion 

Publication of HCF gave rise to two pointless debates, which I will very briefly summarize.

3.1 First pointless debate The first debate, carried out in the pages of Cognition (Pinker & Jackendoff 2004; Fitch, Hauser & Chomsky 2005; Jackendoff & Pinker 2005), limited itself to purely definitional issues: what were the proper contents of FLN and FLB. PJ argued that many more things besides recursion should go into FLN; HCF argued that their limitation of FLN to recursion was a hypothesis not an empirical claim, and the burden of proof lay with those who would extend FLN to include other aspects of language, something they claimed PJ had failed to do. Accordingly the debate was purely definitional; no issues of substance were involved.

3.2 Second pointless debate The second debate, triggered by a sympathetic article in the New Yorker (Colapinto 2007) involved Dan Everett (Everett 2005, 2007) and a number of generativists (see e.g., Nevins, Pesetsky & Rodriguez 2007). Everett. a longtime student of the Piraha language, claimed that Piraha had no recursion, and that therefore recursion could not form part of universal grammar (and maybe, if FLN was just recursion, then there was NO universal grammar.) His opponents insisted that he had misanalysed his data and that Piraha did indeed have recursion. Both sides entirely missed the point that while a biological capacity enables behaviors, it does not enforce them. The absence of recursion from Piraha grammar says no more about universal grammar than the absence of prenasalized consonants or verb serialization from English grammar. In neither debate did anyone question the status of recursion as central to FLN, let alone whether or not recursion really was a language process.

4. The birth of recursion in premature analyses So where does the idea of recursion come from? The idea that syntax is a recursive process originated in early forms of generative grammar, but quickly came to be accepted by everyone. It seemed so self-evident that it has never yet, to my knowledge, been questioned. The idea arose initially from the analysis in Chomsky (1957). At this time, his theory was known as “Transformational-generative grammar” and since transformations formed the most novel (and to many the most salient) aspect of it, it was widely referred to as “Transformational grammar” tout court. The grammar however was divided into two components, phrase structure and transformations. Phrase-structures

 Derek Bickerton

were supplied only for simple sentences, leaving complex sentences to be built out of these by means of the transformational component. Phrase structures were derived from a series of “re-write rules”, which produced strings of abstract symbols consisting of category labels, S(entence), N(oun) P(hrase), V(erb) P(hrase), N(oun), V(erb), P(reposition) etc. Rewrite rules included:

S → NPˉVP NP → (Det)ˉN VP → Vˉ(NP)ˉ(PP) PP → PˉNP

Strings that provided descriptions of simple sentences then served as input to the transformational component. However, for heuristic purposes the operations were frequently described as if they operated on real (surface structure) sentences. Thus “The man you saw yesterday is Harry’s brother” might be described as being produced by insertion of “You saw the man yesterday” into “The man is Harry’s brother” to yield “The man [you saw (the man) yesterday] is Harry’s brother” with subsequent deletion of the repeated “the man”. Thus the Syntactic Structures model involved recursion only in the transformational component, when one prefabricated S was inserted in another prefabricated S. However, this picture was changed radically in Chomsky (1965). The new model introduced “generalized phrase markers”; so that complex sentences were now generated directly by means of expanded rewrite rules. Consequently, recursion was no longer seen as part of the transformational component but formed a core element of phrase structure:

S → NPˉVP NP → (Det)ˉNˉ(PP)ˉ(S) VP → Vˉ(NP)ˉ(PP)ˉ(S)

(The second rule above generates relative clauses, the third generates complement clauses — in both cases referred to as “embedded” sentences.) Consequently “the man you saw yesterday is Harry’s brother” would be generated from the generalized phrasemarker S[NP[Det N S[NP VP]] VP[V NP[N NP[N]]]] which featured one case of S within S and two cases of NP within NP. Accordingly both S-within-S and NP-within-NP seemed to constitute clear cases of recursion. Note, however that recursion is now deduced from a post-hoc, static description and no longer assumed to form part of any sentence-building process. This might already have made recursion look dubious as a process that humans had to execute in order to evolve language. But at this point, of course, a quarter century had to elapse before linguists could even bring themselves to think about evolution.

Recursion 

5. Recursion lingers on while the theory changes Subsequent changes would make generative theory differ even more radically from its beginnings. Transformations continued to be reduced in number, being replaced by interacting principles that achieved similar results at less cost, until finally there was only one left (“Move alpha”). With the arrival of the Minimalist Program, the deep-structure/surface-structure dichotomy gave way to a single structural level with two interfaces, the phonological and the semantic (“logical form”). Processes were reduced to two (“Move” and “Merge”, with attempts to reduce the former to a special case of the latter). “Merge” “takes a pair of syntactic objects and replaces them by a new combined syntactic object” (Chomsky 1995, 226). Whether or not any two such objects can be merged depended on “feature-checking” (determining whether properties and dependencies of objects to be merged matched one another). Merge seems not to have been devised as a description of how sentences are actually produced, but it could serve as such; the process of linking words with one another successively is something that a primate brain, once equipped with a large lexicon, should be able to do with little change beyond some additional wiring. The process is derivational not representational: that is to say it builds structures from scratch, bottom up, rather than starting with a completed string of category labels. It has no preconceived structure: the complex structures of X-bar theory, projecting triple layers of X, X-bar, XP, is abandoned. Its trees consist exclusively of binary branching: ternary branching is excluded, since nodes can have only one sister, and non-branching nodes are excluded because they cannot, by definition, result from applications of Merge.

6. Deriving complexity via Merge Accordingly, let us derive “The man you saw yesterday is Harry’s brother” via Merge: • saw + e → [saw e] Harry’s + brother → [Harry’s brother] (e represents the empty category to be interpreted as co-referential with “man”) • [saw e] + yesterday → [[saw e] yesterday] • is + [Harry’s brother] → [is [Harry’s brother]] • you + [[saw e] yesterday] → [you [[saw e] yesterday]] • man + [you [[saw e] yesterday]] → [man [you [[saw e] yesterday]]] • The + [man [you [[saw e] yesterday]]] → [the [man [you [[saw e] yesterday]]]] • [the [man [you [[saw e] yesterday]]]] + [is [Harry’s brother]] → [[the [man [you [[saw e] yesterday]]]]] [is [Harry’s brother]]]

 Derek Bickerton

Where’s the recursion? We have constructed the sentence by means, not of a recursive, but of an iterative procedure, consisting of repeated applications of an identical process. What is true for relative clauses is equally true for complement ckauses: “Bill thinks that Mary said that John liked her.” • liked + her → [liked her] • John + [liked her] → [John + [liked her]] • that + [John + [liked her]] → [that [John [liked her]]] • said + [that [John [liked her]]] → [said [that [John [liked her]]]] • Mary + [said [that [John [liked her]]]] → [Mary [said [that [John [liked her]]]]] • that + [Mary [said [that [John [liked her]]]]] → [that + [Mary [said [that [John [liked her]]]]]] • thinks + [that [Mary [said [that [John [liked her]]]]] → [thinks [that [Mary [said [that [John [liked her]]]]]]] • Bill + [thinks [that [Mary [said [that [John [liked her]]]]]] → [Bill [thinks [that [Mary [said [that [John [liked her]]]]]]]] Again there is no case of recursion as it is normally defined The irony is that Chomsky is the sole person responsible both for the appearance and disappearance of recursion. His 1957 analysis, created the notion that syntax required recursion. Hs 1995 analysis removed the necessity for assuming recursion. So how is it that Chomsky in HCF is still proposing recursion as the central, perhaps sole content of FLN?

7. Recursion versus iteration Let’s look again at the definition of recursion in HCF a. “..[Recursion] provid(es) the capacity to generate an infinite range of expressions from a finite set of elements…” b. “All approaches agree that a core property of FLN is recursion, attributed to narrow syntax in the conception just outlined. FLN takes a finite set of elements and yields a potentially infinite array of discrete expressions.” It’s worth noting that both definitions avoid any reference to the insertion of syntactic objects into other syntactic objects of the same class. And, as we have seen, Merge is in fact an iterative not a recursive process. Why didn’t HCF bite the bullet and replace “recursion” with “iteration”? I think the reason can only be that iteration alone cannot generate “infinite arrays of discrete expressions”. Iteration of the numbers 1–9 produces no “discrete expressions’

Recursion 

but just a string of unrelated numbers (387964421765988…). Only an additional process coupled with iteration can do this. If we add multiplication to iteration, we can indeed generate an “infinite array of finite descriptions” 5 x 7 = 35 35 x 2 = 70 2 x 9 = 18 18 x 70 = 1360 9 x 7 = 54…..And so on, ad infinitum. What process could one add to iteration to produce such an array in language? The answer lies in the difference between words and numbers. Numbers have no dependencies. Each number (like an animal call, incidentally) is complete in itself and has no special relations, negative or positive, with any other number. Words, to the contrary, have dependencies. If I utter the word “leopard” in isolation, with no expressive intonation, you would know that I was making some kind of reference to an African animal, but you would not know if I was warning you about a leopard, or asking if you had seen one, or denying that there were any around, or merely listing major predators. “Leopard” has to have other words with it if it is to mean anything significant. There has, probably, to be a verb of which it is subject or object. But it cannot be the subject of just any verb; it can be subject of “run” or “kill”, but not of “sing’ or “rust” or “dissolve”. In turn, if we started with “dissolve”, its subject could not be “leopard” or “clock”; it could be “clouds” but not “cloud”, since “dissolve” does not agree with singular nouns in number. Thus the dependencies of words result from their properties, and those properties may be semantic, categorial or grammatical (most times, all three). Indeed, as shown by the feature-checking process in the minimalist program, the iterative procedure in Merge has to proceed along with the process of satisfying the requirements of the various words that are merged: (e.g., liked = Vtrans = requires object; her = 3rd pers. fem. sing. acc. = possible object; liked her = predicate requiring subject; Mary = proper noun, no case = possible subject, and so on.)

8. Why Chomsky can’t jettison recursion So why didn’t HCF simply say that FLN consisted of iteration plus the satisfaction of lexical requirements? Why have generativists now retreated to a definition of a recursive procedure as “one that can indefinitely reapply to its own output, giving rise to a hierarchical structure” (Rizzi 2009)? Because iteration, unlike recursion, cannot be described as a process required only by language. Iteration is a process that lies within the capacity of a wide range of species. Indeed, Rizzi’s definition is one that would be satisfied by a bird building a nest: i. Weaveˉtwoˉtwigsˉtogether. ii. Takeˉaˉthirdˉtwigˉandˉinterweaveˉitˉwithˉtheˉfirstˉtwo. iii. Takeˉaˉfourthˉtwigˉandˉinterweaveˉitˉwithˉtheˉthreeˉalreadyˉwoven.

 Derek Bickerton

and so on. In consequence, generativists are caught in the following dilemma:

A. Adhereˉtoˉtheˉoldˉdefinitionˉofˉrecursionˉwhichˉisˉabsurdˉinˉlight ofˉhowˉ Mergeˉisˉsupposedˉtoˉoperate. B. Adoptˉtheˉnewer,ˉmoreˉinclusiveˉdefinitionˉandˉadmitˉthatˉthisˉis farˉfromˉuniqueˉtoˉhumans.

Note that if (B) is adopted, either FLN would be void or it would consist solely of a lexicon and its requirements. However, Chomsky since the beginning of his career had been wholly committed to the idea that the central part of language is syntax. His compromise with Hauser would not have worked if he had been forced to abandon the centrality of syntax. To preserve that, FLN had to be retained and the content of FLN had to be syntactic not lexical. These goals could be achieved only by appealing to a process that was almost universally supposed to operate in syntax, recursion, even though the most recent developments in Chomsky’s own theory showed that the generation of even the most complex sentences did not require it.

9. How (and why) complexity evolved A more rational response would be to adopt an altogether different model of language evolution. Such a model would claim that, given the kind of lexicon typical of any human language, a purely iterative process that fulfilled the requirements of that lexicon would suffice for the development of structure to whatever level of complexity the language might require. A language might, for reasons of its own, develop only a very low level of complexity, as has been claimed for Piraha, but essentially similar mechanisms would be in play, and nothing in language itself would preclude developing higher levels. The apparent fitting of one structural element (NP or S) inside another of the same type is simply epiphenomenal, arising from the fact that (other than those imposed by individual lexical items) there are absolutely no restrictions on iterative process that generates sentences. Does this mean that there is no unique biological basis for language, no universal grammar? Certainly not. Following Deacon (1997), we can assert that symbolic units are unique to humans and that aspects of the lexicon are genuine universals. After all, the theta-grids of verbs appear to be universal; we know that if we meet a verb in some hitherto unknown language that translates as “sleep”, it will take a single argument, while one that translates as “crush” will take two and one that translates as “tell” will take three. Other things that do not appear to require learning include the rules that determine the reference of empty categories; indeed, since these have no

Recursion 

physically-perceptible expression, it is unclear how, even in principle, they could ever be learned. And we have as supporting evidence the fact that no other species can acquire a human language. Clearly some kind of universal grammar is required for the production of complex sentences. But there is no real evidence that any truly recursive process need be included in that grammar. Rather than the unique content of FLN, recursion in language appears to be no more than an artifact of analysis.

10. Consequences for this conference But if that is the case, what are the implications for what we have all been discussing? It would seem that initially at least, recursion and complexity have been seen as inextricably intertwined. According to Givón (2008), “What makes the syntactic structure of human language complex, in the sense we intend here, is the embedding of clauses in a subordinate — hierarchically lower — position inside other clauses, yielding recursive structure. That is, a node of the same type recurs under a node of the same type. With recursive clause nodes [S] in natural language, such embedding may be found inside a subject or object Noun Phrase (NP), most commonly yielding a subordinate Relative Clause…But embedding and thus recursivity can also occur inside the Verb Phrase (VP), most typically yielding a subordinate Verb Complement” (original emphasis).” If, however, Merge is an iterative process, with no constraints on what can be merged (except those imposed by particular lexical items, which apply solely at each individual attachment — Merge has neither memory nor foresight) then what becomes of the widely-held belief that sentences can be divided into three classes that can in turn be regarded as constituting three stages in three distinct areas: language acquisition, language diachrony, and language evolution? i. combination of words into simple clauses ii. combination of clauses into concatenations of coherent chains iii. condensation of coordinated clauses into tight subordinations Let’s consider the implications for each stage separately, then all together.

10.1 Merge and simple clauses versus no-merge Simple clauses are hierarchically structured. In “John left Sally”, for example, “left” has to be merged with “Sally” before “John” can be merged with “left Sally”. This can be shown by the fact that while material may be attached after the merger of “left”

 Derek Bickerton

and ”Sally”, nothing can be attached to “left” or “Sally” before the two are attached to one another:

JohnˉoccasionallyˉleftˉSally JohnˉleftˉSallyˉoccasionally *JohnˉleftˉoccasionallyˉSally JohnˉwithoutˉmoreˉadoˉleftˉSally *JonˉleftˉwithoutˉmoreˉadoˉSally

In other words, any true simple clause results from application of Merge, not from a beadlike stringing of words based solely on semantic content. The regular word order of true simple clauses is simply an epiphenomenon of the process. In other words, it is superfluous to assume the existence of any form of PS rule: “left” requires some object or person to be left and some person or object to do the leaving, and S → NP VP, VP → V NP is simply a roundabout way of describing what actually happens. Note that in pidgins, early child speech and natural language above sentence level, Merge does not apply, although the reasons why it doesn’t apply are different in all three cases. In pidgins, speakers are able to apply Merge in their own languages because they are fully aware of the requirements and dependencies of words in their own language. However, when confronted with isolated words from languages they don’t know, they are lost. They may assume that the requirements and dependencies of these words are the same as those of their native language, and thus produce the substratuminfluenced, literal-translation-type speech found in some (but by no means all) pidgin speakers. However, even if they choose such strategies, they cannot fully implement them, because virtually all the new items they encounter are lexical not grammatical items, and languages need to merge both types; moreover, in the early stages of a pidgin, even lexical items are sparse, there will inevitably be many gaps, and what words there are do not necessarily “come to mind” at the time needed. Thus pidgin speakers tend to produce only short, structureless utterances without any consistent word order. Very young children, unlike slightly older children and adult pidgin speakers, may not have Merge at all. Alternatively, it is possible that they do have Merge but don’t yet have enough words to merge. This is an empirical issue that needs to be resolved by careful study of the earliest stages of vocabulary growth and the matching of this growth with both the utterances they actually produce and the utterances that (with the vocabulary of any given stage) they could produce. If they don’t have the right words to merge, how could they merge? That they don’t is shown by typical utterances like “Mommy sock”, which could mean “Mommy, please put my sock on”, or “That is Mommy’s sock” (typically, where words are strung together without Merge, ambiguities result that can only be resolved from context). Natural languages lack Merge above sentence level. Why is this? A simple answer would be “Because units above the sentences are too long.” But this is obviously

Recursion 

false: some multi-sentence paragraphs are shorter than some sentences. In fact, one long and complex sentence COULD be a multi-sentence paragraph. And this is probably the crux of the matter. The choice is stylistic. In the earliest stage of language evolution, one assumes that, lacking anything remotely like a full lexicon, No-Merge was the only option. But once Merge emerged, there were two options, and there always will be.(with consequences to be discussed in subsequent sections.) It remains, of course, true that Merge, like any iterative process (think push-ups) requires more effort than its absence and becomes more onerous on the memory the more frequently it is applied in a single series. A French author once produced a novel that consisted of a single sentence (Ndiaye 1988) but it never made the New York Times Best-Seller List.

10.2 Merge and concatenated clauses Co-ordinate clauses are clauses that are individually constructed by Merge and then concatenated by No-Merge, the beads-on-a-string method. For adult speakers, this is often a stylistic choice. Finally John spoke. “The tide is going out”. Finally John said (that) the tide was going out. The question is, of course, are there languages for which it isn’t a choice, and if so why? If the only syntactic process is Merge, then it becomes hard to see how there could be any developmental or physiological obstacle to applying it across the board in all languages. Could the choice here also be a stylistic one, reinforced by conservative tradition? After all, all that biology does for language is offer it a smorgasbord of choices. Not all languages utilize every capacity that human biology makes available, and while change from concatenation to subordination is a common diachronic development, it is far from being a unique one. As Bowern (2008) points out, “we see no overall trend towards greater complexity, and no overall movement towards syntaxis or hypotaxis from parataxis. Rather, as Dahl (2004) has pointed out in other contexts, we see changes and shifts in form and function, but ones which are governed by discourse as much as emerging from it.” Moreover, in cases where nominalizations take the place of VPembedded clauses, there must presumably have been some kind of verb to nominalize, therefore a clausal complement must have historically preceded a non-clausal one. What one sees, in other words, is just what one perceives with serial word-order: a continuous cycling within the envelope biology provides, driven by purely non-structural factors.

10.3 Merge and complex, “subordinated” structures Accordingly the forms found in relative clauses and complement clauses represent not the final stage in some developmental process found throughout ontogeny, phylogeny and diachrony, but rather options that lie within the scope of anyone equipped with

 Derek Bickerton

Merge but that may or may not be selected by particular languages or particular individuals using the same language (where that language’s selection allows it). Evidence in favor of this belief comes from the acquisition of “embedded” sentences (described in Limber 1973). As Limber showed, a wide variety of such sentences (starting with non-finite complements like “I wanna eat it”, including WH-headed clauses, postverbal “that”-clauses and adverbial clauses, closing with object relatives) all come in during the third year, most of them by the middle of that year — in other words, over a four- or five-month period. Moreover, as Limber points out, the fact that it takes the child even this long to acquire a wide range of complex sentence types has little to do with development per se and a great deal to do with the simple order in which the child acquires the kind of verb that will take sentential complements: “The fact that children use these various verbs in object-complement constructions almost immediately upon using them in any construction should not, upon reflection, be very surprising.” Indeed, if the analysis of this paper is correct, this is what is predicted: as soon as the dependencies of a verb are known, Merge will be applied to it.

11. Conclusion The consequences for the evolution of language are clear. First came words — symbolic units with a definite reference, different in kind from animal calls. Then came a pidginlike stringing together of words. Then came Merge, and once Merge was established, it was simply a question of exploiting to the full an iterative process that could be carried on without limit. The degree to which this capacity was exercised became a matter for language or individual choice. It might be asked why, if Merge is the only process required for complex syntax, other animals that have Merge-like processes in other domains do not employ it in their communication. The answer of Hauser, Chomsky & Fitch (2002) is that such a process in other species “represents a modular system designed for a particular function (e.g., navigation) and impenetrable with respect to other systems. During evolution, [this system] may have become penetrable and domain-general…This change from domain-specific to domaingeneral may have been guided by particular selective pressures, unique to our evolutionary past, or as a consequence (by-product) of other kinds of neural re-organization.” A rather more plausible answer is that other species could not apply Merge to their communication because the units of their communication, in contrast with words, are holistic, non-symbolic, and non-referential (to speak of the “functional reference” of predator alarm calls is to ignore the fact that such calls can be translated as instructions to perform particular actions rather than as naming specific predators). Since they are the equivalent of sentences rather than words, and since each unit is situation-specific and designed less to communicate than to improve the caller’s fitness, no earthly purpose

Recursion 

would be served by concatenating them via Merge or anything else. The only surprising thing is that researchers should continue looking for syntactic precursors in other species when it should be obvious that in principle, no syntactic precursor can exist in the absence of words or word-like units. Syntax, no matter how complex, is simply a function of Lexicon plus Merge, and the “complexity” of complex sentences is in the eye of the beholder, rather than the mind/brain of the speaker.

References Bowern, C. 2008. Defining complexity: Historical reconstruction and Nyulnyulan subordination. Ms. Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: The MIT Press. Chomsky, N. 1995. The Minimalist Program. Cambridge MA: The MIT Press. Colapinto, J. 2007. The puzzling language of an Amazon tribe. The New Yorker, April 16, 2007. Dahl, O. 2004. The Growth and Maintenance of Linguistic Complexity. Amsterdam: John Benjamins. Deacon, T. 1997. The Symbolic Species. New York NY: Norton. Everett, D. 2005. Cultural constraints on grammar and cognition in Pirahã. Current Anthropology 46: 621–46 Everett, D. 2007. Cultural constraints on grammar in Piraha: A reply to Nevins, Pesetsky & Rodrigues. LingBuzz, April 2007. Fitch, T., Hauser, M. & Chomsky, N. 2005. The evolution of the language faculty: Clarifications and implications. Cognition 97: 179–210 Givón, 2008. The Genesis of Syntactic Complexity. Introduction. Amsterdam: John Benjamins. Hauser, M., Chomsky, N. & Fitch, T. 2002. The faculty of language: What is it, who has it, and how did it evolve. Science 198: 1569–79 Jackendoff, R. & Pinker, S. 2005. The nature of the language faculty and its implications for evolution of language. Cognition 97: 211–225. Levy, S. 2007. Becoming recursive. Paper presented at the Conference on Recursion in Human Languages, Bloomington IN, April 2007. Limber, J. 1973. The genesis of complex sentences. In Cognitive Development and the Acquisition of Language T.E. Moore (Ed.), 169–186. New York NY: Academic Press. Ndiaye, M. 1988. Comedie classique. Paris: P.O.L. Nevins, I., Pesetsky, D. & Rodrigues, C. 2007. Piraha exceptionality: A reassessment. LingBuzz, March 2007. Pinker, S. & Jackendoff, R. 2004. The faculty of language: What’s special about it? Cognition 95: 201–236. Parker, A. 2007. Was recursion the key step in the evolution of the human language faculty?’ Paper presented at the Conference on Recursion in Human Languages, Bloomington, IN, April 2007. Philips, C. n.d. Lecture 14: Syntax II. University of Delaware. Rizzi, L. 2009. Some elements of syntactic computation. In Biological foundations and origin of syntax, D. Bickerton & Eors Szathmary (Eds), Cambridge MA: The MIT Press.

Index

A Adams, D.Q. 180 adaptive motivation 13 adjuncts 38 adult interlocutors 339 adverbial prefixes 114 adverb placement 152 aerobic bacteria 525 Afroasiatic 35–36, 45, 47 Aggleton, J.R. 472, 476 Aikhenvald, A. 145, 202 Aissen, J. 295 Akan 97, 99 Akkadian 209–211 demonstrative pronoun 208 relative clauses 207 relativization 209 Allport, A. 464 Alpert, N. 434 Alsina, A. 122, 145 Alvarez, Albert 174, 192 Amha, A. 35–36, 47 Amunts, K. 436, 492 analogy 9, 251 Ancient Greek 45 Anderson, P.A.V. 425 Andrews, A.D. 255, 434, 451–452 Angas 31, 44 animacy effect 447 Apte, M.L. 522 arborization 517 argument structure 103, 151–152 Ariel, M. 262 Aristar, A.R. 209 articulatory organization 411 Aslin, R.N. 410, 502 associative network 261, 265 Athabaskan, Tolowa 90, 113–114 Atkinson 9 attentional capacity 475 auditory organization 410 Austronesian 166, 170, 172, 176–177, 194

nominalizations 170 Austronesian 176–178 automatic processing 464 B Baddeley, A. 440–441, 463 Badre, D. 11 Bahasa Indonesia 169 Bahlmann, J. 493–494, 499, 500 Baldwin, G. 15 Ball, C. 217 Bantu 33 Bar-Hillel, Y. 405–406 Barker, M. 399 Barris, R.W. 476 Bartlett, R. 510 Barton, N.H. 511 Bates, E. 8, 302, 320, 329, 421 Bavin, E.L. 259 Beeman, M. 482 behavioral plasticity 525 language pathway 526 Ben-Shachar, M. 440, 497, 500 Biblical Hebrew 115, 210 Bickerton, D. 5, 8, 391, 406, 424 binding 463 biology 9 Birney, D. 434 Blake, B.J. 202 Bloom, L.P. 7, 8, 263, 491 Blozis, S.A. 439 Blumer, D. 484 Boesch, C. 8 Bohnemeyer, J. 120, 131 Bole 31, 44 Bolinger, Dwight 61, 121 Bommelyn, L. 90, 92, 113 Bookheimer, S. 11 Booth, J.R. 10, 454 Bornkessel, I. 434, 497, 500–501 Borod, J. 484 bound morphology/complex words 13 Bowerman, M. 7–8, 279

Bowern, C. 41, 145, 541 Boycott, B.B. 521 Boye, K. 314 Braine, M.D.S. 279, 419 brain plasticity 526 Brandt, S. 260 Brass, M. 445 Brauer, J. 12, 495, 497, 500–502 Broca’s area 11, 412, 417, 436, 465, 491, 492, 500, 502 motor planning in ventral pathway 478–483 automatization 480 grammatical complexity 480–481 inhibitory structure and feedback control 479–480 left hemisphere for ventral trend 481–482 spreading activation and object memory 482–483 Brown, R. 263, 473 Brucart, J.M. 277, 278, 280, 286, 289 Bruce, L. 130 Budiu, R. 425 Bullinaria, J.A. 409 Bulmer, R. 120 Burns, D. 434 Bush, G. 444 Bussey, T.J. 478 Butt, M. 146, 160 Bybee, J.L. 23, 147–148, 151, 202, 242, 252, 258 Byrne, F. 95 C Cacioppo, J.T. 523 Call, J. 8, 429 Campbell, L. 30, 44, 216 Campos, H. 301

 Index canonical SVC 129 types 130 canonical word order 438 Caplan, D. 10, 433–434, 436, 441, 446–447, 453, 497, 500 Carib 88 Carlson, R. 90, 100, 202–203 Carpenter, P.A. 433–434, 436–437, 439, 452, 464 Carter, A. 329 Caucasian 47 center-embedded structures 407 cephalopod CNS 521 Chadic 31, 44 Chafe, Wallace 67, 82, 120, 123–124, 138, 195, 312, 392 Chang, F. 261 Chapman, R.F. 515 Chase, W. 11 Chein, M. 464 Chen, P. 110 Cheyney, D. 8, 352 Chibchan 108 child-adult adjacent turns 345 child-adult usage communicative context 345 conversation 346, 349 descriptive sense 350 direct manipulative speech-acts 350 direct speech-act 346 epistemic 349 manipulative 345 narrative 346–348 speaker-hearer centered 345 child development adult narrative 350 communicative context 350 deontic-manipulative 350 Chinese 189, 190 chloroplasts 525 Chomsky, N. 2–5, 53, 87, 163, 313, 405–406, 419, 433, 461, 492, 510, 531–535, 538, 542 Christiansen, M.H. 262, 491 chunking and attentional capacity 474–475 Clancy, P.M. 293, 294 Claudi, U. 23, 40 clausal complements 323–324

clausal conjunction 93 clausal coordination 396 clause chaining 3, 83, 93 conjunction 81 clause expansion 26–27 clauses 119–120 clause subordination 27 clause union 82–83, 94 diachronic routes to 93–100 embedding languages 94–95, 97–99 cleft constructions 215–216 Clendon, M. 47 Cloney, R.A. 519 co-evolution 215 cognate object constructions 104–105 cognition and syntactic complexity 9 cognitive complexity 10 control 453 cognitive function 409 cognitive representation 485 Cohen, D. 202 Cole, P. 36, 84, 106 co-lexicalization 82–83, 94–96, 114 Collins, P. 223 combination 312, 351 diachrony 351 ontogeny 351 synthesis 6 vs. expansion 13 communicative context 338 communicative ecology 352 compact SVCs 133 Competition Model 422 complement 71 complementation 396 demonstratives 65 English 53 Mohawk 54 complement clauses 29, 43, 48, 54, 397 complement construction 78 complement proposition 343 complement verbs 398 complex clauses 82 diachrony 82, 312 joint construction 351 complex constructions 79, 279, 305, 474, 479 complex dependency 494

complexity 2–5, 240, 277–279, 283, 284, 294, 295, 306. See also syntactic complexity co-evolution 215 complex linguistic constructions 462 complex modal construction 323 complex modal expressions, cross-turn distribution 341 complex modal interactions 317 complex modal structures, joint constructions 343 complex predicates 152–153 complex predications 114 complex structures 163, 525 complex syntactic processes 497 complex syntax 394, 423, 425, 497–502, 542 complex verb phrases diachrony 82 ontogeny 311–312 paratactic precursors 340 syntactic complexity 340 Comrie, B. 34, 42, 44, 46, 156, 165–167, 176, 201, 278, 295, 297 concatenated clauses 399 concatenation 391, 399, 400 conceptual chunking 451 conceptual grouping 398 condensation 239, 241, 305, 351 diachrony 351 ontogeny 351 conflicting information conflict resolution 443–444 semantic memory 445 working memory 444–445 set switching 443, 446 conflict resolution 454–455 connections 1 connotative binding 474 consolidation ventral stream 414–415 Construction frames 278, 290, 302 contextual relevance 319 Cooke, A. 440

Index  coordinate clauses, integration of 23 coordinate-dependent clause 127 coordinate-dependent verb 126 Corballis, M. 427 core conceptual lexicon 414 corpora 153 corpus analysis 154, 155 Corrêa, L.M.S. 255 corticolimbic pathways 481 corticolimbic recursion consolidation 472–473 Coulson, S. 447 co-verb constructions 106–107 Craig, C. 108 Crain, S. 255–256, 259 Crinion, J. 449 Crockford, C. 8 Croft. W. 39, 120 cross-turn collaboration 345 culturally ritualized conventions 473 cuttlefish (systems neuroplasticity) cephalopod chromatophore 521 chromatophore lobes 520 chromatophore muscles 519 striated muscles 519 Czech 182 D Dai, L. 515 Daneman, M. 464 Danish 183 Darwin, C. 428, 511, 523 Dasinger, L. 254 Davies, J. 130, 139 Deacon, T. 15, 417, 538 DeBlois 352 decategorialization 27 Declerck, R. 233–234 Delancey, S. 111, 203 Dell’Orletta, F. 421 demonstrative pronoun 209 deontic 313, 316–317 deontic-manipulative 325 Derrfuss, J. 444–445 desemanticization 27, 75 Desmet, T. 439 D’Esposito, M. 442

Deutscher, G. 25, 33, 39, 41, 166, 205, 208–209 de Villiers, J.G. 259 DevLex maps 417 de Vries 500 diachronic pathways 81, 116 diachronic syntax 231 diachrony 5, 7, 9, 103 complex verb phrases 82 Dick, F. 436 Diessel, H. 24, 65–66, 72, 251–254, 257–259, 277–279, 290, 295, 301, 306, 311–315, 323–324, 329, 332, 339, 345, 351–352, 392 direct epistemic speech-acts 350 direct speech-acts 311, 332 deontic 334 directing attention 337 epistemic 334 epistemic quantification 337 descriptive use 339–340 distributed complexity 340 distributed lexicon 414 distributive network 14 Dittenbach, M. 414 Dittmar, M. 502 diversity 9 Dixon, R.M.W. 323 Donald, M. 427 dorsal limbic circuit 476 dorsal stream 424 Dosenbach, N.U. 445 down-shifting 329 Du Bois, J. 123, 218, 293, 294 Dulcis, D. 518 Dutch 63 Duve, H. 515 E early child language 279, 289, 295, 305, 306 Echeverría, M. 277, 278 Eddy, W.F. 436 Efik 103 Ejerhed, E. 226 Elman, J.L. 261, 277, 279, 306 embedded clauses 8 embedded structure 453 embedding 3, 83, 94–95, 97–99, 128, 277, 284, 305, 399 complementation 81

subordinate 539 Emerson, M.J. 440 emergence 405–429 emergent regularities 284, 292 Engle, R.W. 443 English 44, 63, 94, 146, 179, 204 relative clauses 59 epistemic 313, 316–317 epistemic-informative 325 Ericsson, A. 11 Ervin-Tripp, S. 7, 312–313 Estonian 30, 44 eukaryote protozoan cell 13–14 potassium channels 512 symbiotic arrangement 14 Evenki 34, 44 event integration 85–86 event integration and clause union 85–86 Everett, D. 5, 53, 391, 453, 533 evolution 9 Ewe 40 exaptation 531 executive function 450 conflicting information 435 neuroimaging 440 working memory 435 executive functions 452, 454–455 expansion 66, 200 nominalization 200–201 expansion (analysis), 6 exteriorization 475 External syntax 284 extreme finite (‘non-embedding’) languages 90–93 F Fadiga, L. 424 Fangerau, H. 436 feature map 428 Fernald, R.D. 14 Fernández-Duque, D. 10, 301 399, 440, 455 Fiebach, C.J. 434, 438, 454 finiteness 82 gradients 97–98 nominalization 86 scalarity 88 finiteness gradients 96–97 finite state grammar 492, 499 Finnish 43, 45

 Index Fitch T. 53, 163 Fitz, H. 261 Fodor, J. 409 Fogassi, L. 424 Foley, W. 130, 145, 196 Formosan 171 formulaic systems 392 Fox, B.A. 258, 266, 447 Frajzyngier, Z. 27, 31 Frazier, L. 494 Freedman, M. 483 French 205 Friederici, A.D. 12, 408, 417, 433–434, 437–438, 447, 492–495, 497, 499–502 Friedman, J. 440 Frisch, S. 12 G Gallese, V. 424 gapped evolution 6 Gaston, M. 520–521 Genetti, C. 202 Gennari, S.P. 262, 455 Gentner, D. 251 German 184, 221, 227 Germanic 115, 182, 185, 204, 211–212, 247 Gernsbacher, M.A. 482 Geschwind 11 Gibson, E. 255, 262, 421, 423, 434–435, 439, 448, 453, 455, 466, 494 Gildea, S. 88 Giorgio, A. 494 Giurfa, M. 428 Givón, T. 29, 119, 243, 278, 279, 284, 286, 289, 293–295, 298, 305, 392–393, 480, 539 Gluckman, A. 427, 429 Gobet 11 Goel 449 Gołąb, Z. 196 Goldberg, A.E. 261, 476–477, 479, 484 Goldin, A.L. 512 Goldman-Eisler, F. 124 Goodglass, H. 483 Goodluck, H. 256 Gordon, P.C. 449 Grace, George 119, 392 gradual adaptation 525 graduality 12

grammatical change 25, 43 grammatical development 273 grammatical devices 334 grammatical evolution 239 grammaticalised SVCs 129, 133 grammaticalization 23, 38–39, 43, 82–83 grammatical relations 84 Greek 180 Greenberg, J.P. 202, 512 Gricean principle 140 Grodner, D. 439 Grodzinsky, Y. 12, 408, 417, 433–434, 453, 493–494, 497, 501 Grossman, M. 440, 442, 449 Guenther, F. 411 Guillery, R.W. 471 Günthner, S. 221, 230 H Hagoort, P. 10, 494 Hahne, A. 495 Hale, K. 97, 101, 108, 278 Halford, G.S. 434, 451, 466 Halsband 476 Hamburger, H. 255–256, 259 Hamilton, A.C. 444 Hanging relatives 298, 305 Hanlon, R.T. 519–520 Hannus, A. 464 Harbert, W. 211 Harder, P. 314 Harris, Z.S. 30, 44, 87, 216, 419 Haspelmath, M. 30 Hausa 31, 45 Hauser, Marc 53 Hawkins, J.A. 255 Hayes, A.E. 449 Hazeltine, E. 444 heavy verbs 105 Heine, B. 6–8, 23, 27, 39, 40, 42, 64–65, 75, 114–115, 147, 165–166, 200–201, 206–208, 216, 314, 323, 352 Hendler, J.A. 472 Hermon, G. 254 Herrick, C.J. 475 heteromodal association cortex 467 hierarchic organization 2 hierarchical structure 453

hierarchical syntactic structure 399 hierarchic systems 1, 2 Hille, B. 511–512, 514 Hilpert, M. 226 Hippocrates 509 Hitch, G. 463 Hochstadt, J. 450 Hock, H.H. 211 Hockett, C.F. 243 Hodgkin, A.L. 515 Hofland, K. 218 Hofstadter, D. 251 Holm, J.A. 40 Holmes, V.M. 439, 519 Hopper, P.J. 23, 25, 42, 146–147, 151, 165, 201, 204, 207, 220, 223, 258, 314, 352 Horie, K. 167 Howerter, A. 440 Huallaga Quechua 179 Huang, C. 173, 187 Huddleston, R. 270 Hünnemeyer, F. 23, 40 Hurford, J. 195, 424 hypotaxis 165 I Icelandic 211 ideophones 106 Igbo 103 Ijo 100 Ik 45 incorporated objects 111 independent conversational turn 286 independent verb 126 indirect speech-acts 331 Indonesian 191 inferior frontal gyrus 417, 436 inflectional morphology 242 inflectional systems 241 inhibitory specification 461–462, 479 Internal syntax 284, 294, 303 intonation 53, 58 intonational boundaries 449 intonational contour 73 intonational packaging 101 intonation contour 82 intonation unit 68 Irish 45 Iroquoian 54

Index  item-based patterns 419–424 iteration 429 J Jackendoff, R. 53, 120, 391, 533 Jackson, J.H. 474 Jan, L.Y. 512 Japanese 166–167, 188–189, 191 Jeannerod, M. 410, 473 Jespersen 204 Johansson, M. 225 joint construction 287 Jones, D.K. 222 Jusczyk, P.W. 408 Just, M.A. 10, 433–437, 439, 454, 464–465 K Kaan, E. 10, 433, 494 Kalam 100, 102, 119–120, 123, 125, 132, 135, 141 Kandel, E.R. 511–512 Kane, M.J. 442 Kanuri 26, 45 Karlsson, F. 466 Karolija, N. 312 Keenan, E.L. 42, 165–167, 176, 255, 278, 295 Keller, T.A. 436–437 Kerslake, C. 48 Khoisan 29, 38, 45 Khwe 45 K’ichee’, 186 Kidd, E. 255, 259, 263, 266 Kim, Y.-J. 222, 234–235 King, J. 439, 453–454, 464 Kintsch, W. 425 Kirby, S. 491 Kliegl, R. 421 Kluender, R. 439 Kluge, F. 204 Knoblich, G. 410, 424 Ko, K. 439 Kobon 141 Koelsch, S. 449 Kohonen, T. 410 König, C. 45 Koops, C. 220, 236 Kordofanian 35 Koromfe 45 Kral, A. 518 Krongo 35, 46 Kuczaj, S. 263

Kuhl, P.K. 410 Kuno, S. 298 Küntay, A. 312 Kutas, M. 439, 453–454 Kuteva, T. 6–8, 23, 26–27, 27, 39, 42, 64, 75, 114–115, 147, 165–166, 200, 206–208, 314, 323 Kwa 103 L Labov, W. 135, 139 Lacerda, F. 411 Lahiri, A. 146, 160 Lahu 187–190 Lakoff, G. 415 Lambrecht, K. 216–217, 232, 236, 253, 278, 293 Lane, J. 120, 129, 130, 142 Langacker, R.W. 42 language diachrony 6 language ontogeny 6 La Polla, R. 295 Larkin, W. 434 Larsen, T.W. 186 Lashley, K.S. 474 lateral neurosecretory cells 517 Latin 46 Latin-derived prepositional verbs, English 109 Latin subjunctive 195 Laughter 523 Laz 36, 47 learning circuits 471–472 Lebel, C. 494 Lefcourt, H.M. 522 left anterior negativity 448 Lehmann, C. 195, 212, 240 Lenci, A. 421 Levelt, W.J.M. 409 Levine, R.B. 518 Levy, S. 532 Lewis, G.L. 175 Lewis, R.L. 448 lexicalization 116 lexical nominalizations 92 lexical organization 412–414 lexical-semantics 11 lexical-specific structural frames 279 lexicon 413 Lexis 240, 244–246 Li, C.N. 47, 191, 412

Lieven, E. 252, 278, 301, 493 Lightfoot, D.W. 151 Limber, J. 279, 542 Lindenberg, R. 436 Linell, P. 312 linguistic complexity 415, 484 articulatory organization 411–412 auditory organization 410–411 consolidation 414–416 Liotti, M. 478, 481 Lipinski, E. 210 Liven, E. 278, 301 local phrase structures 494 local syntactic processes 497 Lohmann, G. 438 Loi, P.K. 515, 519 Lope Blanch, J.M. 298 Luu, P. 462, 468, 470, 472, 477, 480 Lyons, J. 415 M Maale 35–36, 47 MacDonald, M.C. 262, 423 MacWhinney, Brian 7, 13, 263, 311, 314, 409, 412, 415–416, 418–419, 420, 421–424, 426, 428 Maddrell, S.H.P. 515 Mak, W.M. 262, 447 Malayo-Polynesian 171 Malle, B. 399 Mandarin 39, 47 manipulation verbs 318, 334 Manjiljarra 47 Marcus, G. 410 Margulis, L. 14 Matisoff, J.A. 187–191, 196, 209 matrix clause 78 Mayr, Ernst 13, 15 Mayrinax Atayal 171 McCandliss, B.D. 444 McGhee, P.E. 523 McMillan, C. 440 McWhorter, J.H. 242 medial verbs 126 Median, J. 251 Meillet, A. 241 Meltzoff, A.N. 352 memory consolidation 466–474

 Index memory consolidation limbic cortex 461 neocortex 461 mental models 424–427 mental set switching 454 Menzel, R. 428 merge 408, 535, 539–542, 544 stylistic 541 Merkl, D. 414 Messenger, J.B. 520 Mesulam, M.M. 467 metamorphosis 517 Meyer, D.E. 482 Meyer-Lübke, W. 205 Middle English 185 Milkman, R. 511 Miller, J. 220, 226, 410, 433, 463 minimalism 531 Miranda, E.R. 411 “mirror” neurons 424 Mishkin, M. 472, 476 Miskitu 97, 101 Mitchell, B. 211 Mithun, M. 43, 53, 55, 101 mitochondria 525 mitosis 14 Miyake, A. 433–434, 440 modal expansion 343 modal intention 320, 322 modal interaction 325 modal interaction units 315 modality verbs 318, 334 modal negotiations 311 modal reinforcement 343 modal structure, Theories of Mind 352 modern English 185 Mohawk 55, 61, 63, 65, 69, 72, 75, 77 complement constructions 70 definite article 72 pronominal prefixes 54 Monsell, S. 445 Montemagni, S. 421 Moore, P. 440 morphological causatives 84 morphological complexity 7 morphosyntax expansion 26–27 Morris, J. 439 move 535

Mulac, A. 314 Muller, K. 439 Munske, H.H. 204 Mushiake 476 N Nadel, L. 471 narratives 135 narrative SVC 122, 134, 136, 138–139 Ndiaye, M. 541 Neafsey, E.J. 472–473, 476 neural plasticity 509 cellular 514–518 cognitive (laughter), 522–525 invertebrates 521 molecular (sodium channels in eukaryotes), 510–514 systems (cuttlefish), 518–522 neuroplasticity 526 Neville, H.J. 495 Newman, P. 31, 45 Newport, E. 279, 410 Niger-Congo 29, 33, 40, 45, 97, 103 Nilo-Saharan 45 Nilsson, B. 244 No. Paiute 111 nodes 2 “nominalization marker,” 169 nominalizations 170–176, 186–187, 194, 199, 201 back-formation 204 expansion 199–200 integration Akkadian relative clauses 207–210 subordination 199 nominalizing (embedding) languages 88 nominal objects 93, 323 nominal properties 28 non-displaced reference 345 non-linearity 241 non-subject relatives 252, 266, 273 Noonan, M. 28, 34, 36, 39, 45, 47, 48, 55, 195 Norman, W.N. 186 Northern-Iroquoian 62 Novick, J.M. 446 Numic 88

O Oberecker, R. 495 object 38 object extraction 153 object-extracted relative 433 ambiguity 434 perspective shifts 435 storage cost 434 syntactic dependencies 439 syntactic integration 435 working memory 439 object memory 482 object relatives 260, 262 object-relative sentences 465 Object rels 301 object representations 472 Ochs, E. 8, 311–313 O’Grady, W. 255, 423 Old English 146, 211 Oller, D.K. 408, 411 Olson, M. 130, 145 ontogeny 5, 9, 495, 502 orbital frontal region 476 O’Regan, J.K. 439 organisms/systems, complexity 1 Osam, E.K. 97, 99, 104 P Packard, M.G. 449, 519 Pagliuca, W. 242 Palacios, M. 298 Pancor Ngeno-Ngené Sasak 167 Pandya, D.N. 466–467, 474, 476–477, 483, 492 Papez, J.W. 471 Papiamentu 40 parataxis 165, 240, 244, 247, 351 Pardeshi, P. 160 Parker, A. 532 Parkinson disease 449 participial nominalizations 196 Passingham, R.E. 476 Pawley, A. 67–68, 94, 102, 114, 120, 124, 129, 130, 176, 177 Pearson, T.R. 318 Peelle, J.E. 440 Pepperberg, I.M. 8 perception-cognitionutterance verbs 334

Index  perirhinal cortex 476 Perkins, R.D. 242 Perlmutter 423 Persian 47 perspective and gesture 427–429 perspective shift 425–426 Pesetsky, D. 533 Petersen 11 Petrides, M. 492 Phillips, S. 434 phrase structure grammar 492, 499 phrase structures 534 phylogeny 5, 502 Piaget, J. 411 pidgin 540 Pinker, S. 53, 391, 409, 491, 533 Pirrelli, V. 421 Pisin, Tok 132 Pittner, K. 211 Pléh, C. 423 Popper, Karl 15 positional patterns 418–419 Posner, M.I. 11, 12, 440, 444 Post, M.W. 206 Postle, B.R. 444 post-verbal incorporated prepositions, in English 110 Povinelli, D.J. 352 pragmatic accommodation 232, 236 conventionalization 236 Prat, C.S. 465 predicate 38 Premack 352 presuppositional structure 232 pre-verbal incorporation 108 Romance and Germanic 109 Pribram, K.H. 476 primary visual areas 467 Prince, E. 217, 233 Prinz, W. 410 prokaryotes 511 potassium channels 512, 514 pronominal prefixes 54 prosodic intonation 58 Mohawk 57 syntactic structures 61 prosodic contour 60, 67 prosodic features 414

prosodic patterns 53, 60, 78 prosodic phrase 58 prosodic phrase/intonation unit 78 prosodic structures 61, 63, 77 prosody 57, 79 complement constructions 71 demonstrative constructions 72 Mohawk 57 prototype 87 prototypical transitive clauses 273 Provine, R.R. 523 pseudoclefts 215, 217–218, 220, 227, 231 discourse frequency of 227–228 English 218–219, 232 wh-clause 220–225 frequency 219 German 227, 231 Germanic 231 grammaticalization 231 presuppositional structure 232–233 relative clause 228–229 Swedish 227, 231 wh-clause 230–231, 233–234 pseudo-coordination 145, 148, 156, 160 psycholinguistic complexity 415 Pujol, J. 493, 502 Pullum, G.K. 270 Pulvermüller, F. 12 Purves, D. 518 Q Qiang 173 Quechua 36, 46, 47 Quechuan 88 R Raichle, M. 11 Rama 108 Ranganathan, R. 511 Rauber, A. 414 Reali, F. 262 recursion 163, 406, 428, 429, 461, 532–533, 536 Chomsky 537–538

complexity 531, 533, 535–536, 538–539 evolution 534 vs. iteration 536–537 recursive 194 processing 485 rules 165 recursivity 4 Reed, C.M. 519 reentrant connections 485 Reesink, G. 130 reference tracking 279, 289 referential coherence 279 referential competition 286 referential continuity 299 referential integration 85 Reh, M. 35, 216 Reid, L.A. 176 reinforcement 345 relational grammar 82 relative clauses 3, 164, 167, 189–190, 195, 207, 212, 252, 254, 271, 446, 539 animacy 264 English 256–257, 259, 263 frequency 251, 273 German 228, 256–257, 260, 263 non-subject relatives 268, 272 similarity 251 subject relatives 268, 272 relativization 73, 166, 211, 396 demonstratives 73 English 56 Mohawk 56 relative constructions 277, 279 Relative pronoun 277, 280, 281 relativization 166–170, 278, 297, 298, 305 Rennison, J.R. 45 representation 463 representational actions 1, 119, 215 restrictive relative construction 78 restrictive relatives 77 Resumptive linking 296 re-write rules 534 Rice, K. 114 Rilling, J.K. 502 Rizzi, L. 537 Rizzolatti, G. 424, 492 Roberts, J. 139

 Index Roder, B. 500 Rodriguez 533 Rojas-Nieto, C. 277–280, 392 Roland, D. 455 Romance 181 Ross, J.R. 122, 176 Ross-Hagebaum, S. 220 Ruch, W. 523 Rumbaugh, D.M. 8 Russian 181 S Sachs, J. 263 Saffran, J.R. 410 Sanides, F. 492 Santi, A. 501 Sarah 263 Saramaccan 95 Sasak 168–170, 189, 194 nominalization markers 171 Savage-Rumbaugh, S. 8 scalarity 88 Schlesewsky, M. 434, 438, 501 Schmahmann, J.D. 12, 493, 502 Schneider, W. 11, 464 Schütz-Bosbach, S. 410 Schvaneveldt, R.W. 482 Scollon, R. 7, 8, 312, 313 segmentation 451 Seidenberg, M.S. 279, 423 Seitz, R.J. 436 semantic bleaching 151 semantic deficits 473 semantic memory 473–474 semitic 207, 210 Senufu 90 sequence analysis 416 sequence detection 428 serial-verb languages 95–96, 99–100 serial verb constructions 119–125, 128 compact 130–134 narrative 134–141 Kalam 125–129 embedding 128 serial verb constructions 128 verbal clauses 126 serial verb languages 95–96, 99–100 serial verbs 93

set switching 446, 449, 455 Seyfarth, R. 11, 352 Sheldon, A. 255 Sher, P.K. 523 Sherman, S.M. 471 Shibatani, M. 84, 406 Shiffrin, R.M. 9, 11, 464 Shima, K. 476, 477 Silva-Pereyra, J. 495 Silverstein, M. 42 Simon, H. 2, 11 Slavic 181 Sleigh, M. 14 Slobin, D. 6, 15, 252 Smith, M.D. 259, 441 Sneddon, J.N. 187 Snow, C. 314 Snyder, C.R.R. 11 Spanish 99, 181, 188, 190, 192, 277, 279, 280, 295, 298, 301, 304, 305 auxiliaries 96 spatial context 476 spatial deixis 75 spatial integration 85 spatio-temporal displacement 329 speech-act 320 speech-act value 331, 335 declarative or interrogative 331 deontic-manipulative 331 description 336 direct manipulative 335 Squamish 36, 47 Squire, L.R. 461, 470 Starosta, S. 177 starting small 277, 279, 298, 306 Stern, M.B. 450 Stong-Jensen, M. 211 storage 421–423 Stromswold, K. 497, 500 Strong, M. 512, 514 stroop task 449 subject 38 relatives 256, 260, 267, 273 subordinate clauses 26, 61 subordination 165, 199, 400 superior temporal gyrus 491, 493 Suppes, P. 263 Supyire 95, 100 Swaab, T.Y. 433

Swahili 33 Swedish 149, 184, 227 Swedish corpora 153 pseudo-coordination 103, 145, 148–150 Swinney, D.A. 494 switching mental sets 445 Syder, F. 67, 124 synchronic typology 81 syntactic adjustment 87 syntactically complex sentences 493 syntactically complex structures 399 syntactic complexity 391, 450, 453, 455, 491 cognition and 9 developmental trends 6 embedding 399 informational content 400 memory limitations 464 motivations for complexity 393–394 neural underpinnings 433–455 neuro-cognition of 10 processing load 400 syntactic frame 285, 306 syntactic gap 284, 294, 295, 304 syntacticization (embedding), 116 Syntactic integration 286, 288, 298 syntactic movement 150, 453–454 syntactic nominalization 199 syntactic operations 453 syntactic processing 446 syntactic specificity 440 syntactic structure 61 syntax 242, 532 syntaxis 240, 244–246, 248, 351 T Tagalog 171, 176 Tairora 132 Tallerman, M. 6, 13 Talmy, L. 120 Tavakolian, S.L. 253, 255–256, 259 Teleman 148–150

Index  temporal integration 85 terminal modification 12–13 thalamic projections 467 thalamic resonance 470–472 thematic coherence 319 Thompson, S. 47, 63, 191, 201, 204, 258, 266, 314, 447 Thomson 311 Thornes, T. 111 Thulborn, K.R. 436 Thurman, R. 88 Tibbitts, B. 108 Tibetan/Uto-Aztecan 176 Tibeto-Burman 88, 173, 202, 209 Tok Pisin 100 Tolowa 90–91 finiteness of 92 Tomalin, M. 406 Tomasello, M. 8, 251–254, 257–259, 277–279, 295, 301, 306, 311–314, 329, 332, 345, 351–352, 427, 429, 491, 493 Tomlin, R. 393 Topic continuity 297–299, 301, 305, 306 Toupin, C. 254 transformational grammar 406, 533 transformations 10 transitive clauses 266 Trask, R.L. 203 Traugott, E.C. 23, 25, 42, 146, 165, 201, 207, 218–219, 224, 314, 352 Traxler, M.J. 262, 439 Trueswell, J.C. 262 Truman, J.W. 518 Tswana 106 Tublitz, N. 15, 41, 515, 517, 519–521 Tucker, D.M. 409, 414, 424, 462, 468, 470–472, 477–478, 478, 480–481, 485 Turkish 48, 175, 244

U Underriner, J. 114 unicinate fasciculus 500 usage-based approach 273 Ute 88, 98, 104, 116, 243, 244 nominalizations 112 Uto Aztecan 88, 111, 174 Uzbek 34, 48 V Valenzuela, P. 91 van der Bilt, A. 510 Van Dyke, J.A. 448 Van Valin, R.D. 130 V-complement constructions 311 ventral corticolimbic circuit 462 ventral corticolimbic pathways 476 action regulation 477 ventral limbic pathway 462, 476, 482 ventral stream 424 ventrolimbic networks 472 verb adjacency 94–96 verb adjunct construction 125 verb adjuncts 125 verbal clauses 126 verbal complements 3 verbalization 212 verbal properties 28 verbal working memory 454 verb complement 93, 164, 539 verb-particle construction 110 Verhagen, A. 63 visceromotor function 483 von Cramon 434, 438 W Waddington 15 Wager, T.D. 441, 442, 443, 445, 446 Wagiman 85, 106, 107 Wagner, A.D. 11 Warren, T. 262 Washburn 8

Waters, G.S. 433, 434 Watson, D. 439, 449 Watters, D. 88 Weber, D. 88, 179, 186, 187 Weinert, R. 220 Wellman, H.M. 352 Wernicke’s aphasia 483 Wernicke’s area 11, 408, 465, 483 Westermann, G. 411 Wh-clause 221, 233 English pseudoclefts 223 German 230 pragmatic accommodation 235 pseudocleft 222, 235 Swedish 230 White, S.A. 14 Wiklund, A.-L. 148 Wild, B. 523 Williams, R.S. 439 Williamson, K. 100 Wilson, S. 85, 107, 410, 434 Wingfield, A. 440 Witzki, A.H. 440 Woodruff 352 working memory 440–442, 450, 453, 62–463, 465 Wray, A. 120, 392 X X-bar 423 !Xun 38, 48 Y Yakovlev, P.I. 475 Yaqui 174–175, 177, 189, 194 Yatye 103 Yeung, N. 446 Yoruba 84, 103 Young, J.Z. 108, 521 Z Zelazo, P.D. 452–453 Zhao, X. 412 Zipf, G. 116 Zuberbühler, K. 8 Zubizarreta, M.L. 293

Typological Studies in Language A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 87 Cyffer, Norbert (ed.): Negation Patterns in West African Languages and Beyond. Expected Forthcoming 86 Mahieu, Marc-Antoine and Nicole Tersis (eds.): Variations on Polysynthesis. The Eskaleut languages. 2009. x, 312 pp. 85 Givón, T. and Masayoshi Shibatani (eds.): Syntactic Complexity. Diachrony, acquisition, neurocognition, evolution. 2009. vi, 553 pp. 84 Newman, John (ed.): The Linguistics of Eating and Drinking. 2009. xii, 280 pp. 83 Corrigan, Roberta, Edith A. Moravcsik, Hamid Ouali and Kathleen M. Wheatley (eds.): Formulaic Language. Volume 2. Acquisition, loss, psychological reality, and functional explanations. xxii, 344 pp. + index. Expected May 2009 82 Corrigan, Roberta, Edith A. Moravcsik, Hamid Ouali and Kathleen M. Wheatley (eds.): Formulaic Language. Volume 1. Distribution and historical change. xxii, 295 pp. + index. Expected May 2009 81 Corbett, Greville G. and Michael Noonan (eds.): Case and Grammatical Relations. Studies in honor of Bernard Comrie. 2008. ix, 290 pp. 80 Laury, Ritva (ed.): Crosslinguistic Studies of Clause Combining. The multifunctionality of conjunctions. 2008. xiv, 253 pp. 79 Abraham, Werner and Elisabeth Leiss (eds.): Modality–Aspect Interfaces. Implications and typological solutions. 2008. xxiv, 422 pp. 78 Harrison, K. David, David S. Rood and Arienne Dwyer (eds.): Lessons from Documented Endangered Languages. 2008. vi, 375 pp. 77 Seoane, Elena and María José López-Couso (eds.): Theoretical and Empirical Issues in Grammaticalization. In collaboration with Teresa Fanego. 2008. x, 367 pp. 76 López-Couso, María José and Elena Seoane (eds.): Rethinking Grammaticalization. New perspectives. In collaboration with Teresa Fanego. 2008. x, 355 pp. 75 Frajzyngier, Zygmunt and Erin Shay (eds.): Interaction of Morphology and Syntax. Case studies in Afroasiatic. 2008. v, 234 pp. 74 Kurzon, Dennis and Silvia Adler (eds.): Adpositions. Pragmatic, semantic and syntactic perspectives. 2008. viii, 307 pp. 73 Ansaldo, Umberto, Stephen Matthews and Lisa Lim (eds.): Deconstructing Creole. 2007. xii, 292 pp. 72 Næss, Åshild: Prototypical Transitivity. 2007. x, 240 pp. 71 Nedjalkov, Vladimir P. (ed.): Reciprocal Constructions. With the assistance of Emma Š. Geniušienė and Zlatka Guentchéva. 2007. xxiii, 2219 pp. (5 vols.). 70 Zúñiga, Fernando: Deixis and Alignment. Inverse systems in indigenous languages of the Americas. 2006. xii, 309 pp. 69 Aranovich, Raúl (ed.): Split Auxiliary Systems. A cross-linguistic perspective. 2007. vii, 277 pp. 68 Abraham, Werner and Larisa Leisiö (eds.): Passivization and Typology. Form and function. 2006. x, 553 pp. 67 Veselinova, Ljuba N.: Suppletion in Verb Paradigms. Bits and pieces of the puzzle. 2006. xviii, 236 pp. 66 Hickmann, Maya and Stéphane Robert (eds.): Space in Languages. Linguistic Systems and Cognitive Categories. 2006. x, 362 pp. 65 Tsunoda, Tasaku and Taro Kageyama (eds.): Voice and Grammatical Relations. In Honor of Masayoshi Shibatani. 2006. xviii, 342 pp. 64 Voeltz, F. K. Erhard (ed.): Studies in African Linguistic Typology. 2006. xiv, 426 pp. 63 Filimonova, Elena (ed.): Clusivity. Typology and case studies of the inclusive–exclusive distinction. 2005. xii, 436 pp. 62 Couper-Kuhlen, Elizabeth and Cecilia E. Ford (eds.): Sound Patterns in Interaction. Crosslinguistic studies from conversation. 2004. viii, 406 pp. 61 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 2. 2004. xii, 319 pp. 60 Bhaskararao, Peri and Karumuri Venkata Subbarao (eds.): Non-nominative Subjects. Volume 1. 2004. xii, 325 pp.

Syntactic Complexity: Diachrony, acquisition, neuro-cognition, evolution (Typological Studies in Language)

The genesis of syntactic complexity: diachrony, ontogeny, neuro-cognition, evolution

Adpositions: Pragmatic, Semantic and Syntactic Perspectives (Typological Studies in Language)

Language Evolution and Syntactic Theory

Language Acquisition: Studies in First Language Development

Coordinating Constructions (Typological Studies in Language)

Prototypical Transitivity (Typological Studies in Language)

Investigations In Instructed Second Language Acquisition (Studies on Language Acquisition)

Ideophones (Typological Studies in Language, 44)

The Neurocognition of Language

Coordinating Constructions (Typological Studies in Language)

Dimensions of Possession (Typological Studies in Language)

Ideophones (Typological Studies in Language, 44)

Deconstructing Creole (Typological Studies in Language)

Coordinating Constructions (Typological Studies in Language)

Studies in African Linguistic Typology (Typological Studies in Language)

Creoles, their Substrates, and Language Typology (Typological Studies in Language)

Formulaic Language, Vol. 2: Acquisition, loss, psychological reality, and functional explanations (Typological Studies in Language)

Studies in Evidentiality: v. 54 (Typological Studies in Language)

Language Evolution (Studies in the Evolution of Language)

Language Evolution (Studies in the Evolution of Language)

The Acquisition of Syntactic Knowledge

Benefactives and Malefactives: Typological Perspectives and Case Studies (Typological Studies in Language)

Functional Categories in Learner Language (Studies on Language Acquisition)

Theoretical and Empirical Issues in Grammaticalization (Typological Studies in Language)

New Reflections on Grammaticalization Vol 49 (Typological Studies in Language)

Language Complexity: Typology, contact, change (Studies in Language Companion Series)

A Typology of Purpose Clauses (Typological Studies in Language)

Reflexives: Forms and Functions (Typological Studies in Language)

Case, Typology and Grammar (Typological Studies in Language)

Language Complexity as an Evolving Variable (Studies in the Evolution of Language)

Syntactic Complexity: Diachrony, acquisition, neuro-cognition, evolution (Typological Studies in Language)

The genesis of syntactic complexity: diachrony, ontogeny, neuro-cognition, evolution

Adpositions: Pragmatic, Semantic and Syntactic Perspectives (Typological Studies in Language)

Language Evolution and Syntactic Theory

Language Acquisition: Studies in First Language Development

Coordinating Constructions (Typological Studies in Language)

Prototypical Transitivity (Typological Studies in Language)

Investigations In Instructed Second Language Acquisition (Studies on Language Acquisition)

Ideophones (Typological Studies in Language, 44)

The Neurocognition of Language

Coordinating Constructions (Typological Studies in Language)

Dimensions of Possession (Typological Studies in Language)

Ideophones (Typological Studies in Language, 44)

Deconstructing Creole (Typological Studies in Language)

Coordinating Constructions (Typological Studies in Language)

Studies in African Linguistic Typology (Typological Studies in Language)

Creoles, their Substrates, and Language Typology (Typological Studies in Language)

Formulaic Language, Vol. 2: Acquisition, loss, psychological reality, and functional explanations (Typological Studies in Language)

Studies in Evidentiality: v. 54 (Typological Studies in Language)

Language Evolution (Studies in the Evolution of Language)

Language Evolution (Studies in the Evolution of Language)

The Acquisition of Syntactic Knowledge

Benefactives and Malefactives: Typological Perspectives and Case Studies (Typological Studies in Language)

Functional Categories in Learner Language (Studies on Language Acquisition)

Theoretical and Empirical Issues in Grammaticalization (Typological Studies in Language)

New Reflections on Grammaticalization Vol 49 (Typological Studies in Language)

Language Complexity: Typology, contact, change (Studies in Language Companion Series)

A Typology of Purpose Clauses (Typological Studies in Language)

Reflexives: Forms and Functions (Typological Studies in Language)

Case, Typology and Grammar (Typological Studies in Language)

Language Complexity as an Evolving Variable (Studies in the Evolution of Language)

Recommend Documents